PARTICLE PHYSICS

AND INFLATIONARY COSMOLOGY1

Andrei Linde

Department of Physics, Stanford University, Stanford CA 94305-4060, USA

1
This is the LaTeX version of my book “Particle Physics and Inflationary Cosmology” (Harwood, Chur,
Switzerland, 1990).
Contents

Preface to the Series ix

Introduction x

CHAPTER 1 Overview of Unified Theories of Elementary Particles and the Infla-

tionary Universe Scenario 1
1.1 The scalar field and spontaneous symmetry breaking 1
1.2 Phase transitions in gauge theories 6
1.3 Hot universe theory 9
1.4 Some properties of the Friedmann models 13
1.5 Problems of the standard scenario 16
1.6 A sketch of the development of the inflationary universe sce-
nario 25
1.7 The chaotic inflation scenario 29
1.8 The self-regenerating universe 42
1.9 Summary 49

CHAPTER 2 Scalar Field, Effective Potential, and Spontaneous Symmetry Break-

ing 50
2.1 Classical and quantum scalar fields 50
2.2 Quantum corrections to the effective potential V(ϕ) 53
2.3 The 1/N expansion and the effective potential in the
λϕ4 /N theory 59
2.4 The effective potential and quantum gravitational effects 64

CHAPTER 3 Restoration of Symmetry at High Temperature 67

3.1 Phase transitions in the simplest models with spontaneous
symmetry breaking 67
3.2 Phase transitions in realistic theories of the weak, strong, and
electromagnetic interactions 72
3.3 Higher-order perturbation theory and the infrared
problem in the thermodynamics of gauge fields 74

CHAPTER 4 Phase Transitions in Cold Superdense Matter 78

4.1 Restoration of symmetry in theories with no neutral
currents 78
CONTENTS vii

4.2 Enhancement of symmetry breaking and the

condensation of vector mesons in theories with
neutral currents 79

CHAPTER 5 Tunneling Theory and the Decay of a Metastable Phase in a First-

Order Phase Transition 82
5.1 General theory of the formation of bubbles of a new phase 82
5.2 The thin-wall approximation 86
5.3 Beyond the thin-wall approximation 90

CHAPTER 6 Phase Transitions in a Hot Universe 94

6.1 Phase transitions with symmetry breaking between the weak,
strong, and electromagnetic interactions 94
6.2 Domain walls, strings, and monopoles 99

CHAPTER 7 General Principles of Inflationary Cosmology 108

7.1 Introduction 108
7.2 The inflationary universe and de Sitter space 109
7.3 Quantum fluctuations in the inflationary universe 113
7.4 Tunneling in the inflationary universe 120
7.5 Quantum fluctuations and the generation of adiabatic density
perturbations 126
7.6 Are scale-free adiabatic perturbations sufficient
to produce the observed large scale structure
of the universe? 136
7.7 Isothermal perturbations and adiabatic perturbations
with a nonflat spectrum 139
7.8 Nonperturbative effects: strings, hedgehogs, walls,
bubbles, . . . 145
7.9 Reheating of the universe after inflation 150
7.10 The origin of the baryon asymmetry of the universe 154

CHAPTER 8 The New Inflationary Universe Scenario 160

8.1 Introduction. The old inflationary universe scenario 160
8.2 The Coleman–Weinberg SU(5) theory and the new
inflationary universe scenario (initial simplified version) 162
8.3 Refinement of the new inflationary universe scenario 165
8.4 Primordial inflation in N = 1 supergravity 170
8.5 The Shafi–Vilenkin model 171
8.6 The new inflationary universe scenario: problems and prospects176

CHAPTER 9 The Chaotic Inflation Scenario 179

9.1 Introduction. Basic features of the scenario.
The question of initial conditions 179
CONTENTS viii

9.2 The simplest model based on the SU(5) theory 182

9.3 Chaotic inflation in supergravity 184
9.4 The modified Starobinsky model and the combined
scenario 186
9.5 Inflation in Kaluza–Klein and superstring theories 189

CHAPTER 10 Inflation and Quantum Cosmology 195

10.1 The wave function of the universe 195
10.2 Quantum cosmology and the global structure of the
inflationary universe 207
10.3 The self-regenerating inflationary universe and quantum cos-
mology 213
10.4 The global structure of the inflationary universe and the
problem of the general cosmological singularity 221
10.5 Inflation and the Anthropic Principle 223
10.6 Quantum cosmology and the signature of space-time 232
10.7 The cosmological constant, the Anthropic Principle, and redu-
plication of the universe and life after inflation 234

CONCLUSION 243

REFERENCES 245
Preface to the Series

The series of volumes, Contemporary Concepts in Physics, is addressed to the professional

physicist and to the serious graduate student of physics. The subjects to be covered will
include those at the forefront of current research. It is anticipated that the various volumes
in the series will be rigorous and complete in their treatment, supplying the intellectual
tools necessary for the appreciation of the present status of the areas under consideration
and providing the framework upon which future developments may be based.
Introduction

With the invention and development of unified gauge theories of weak and electromag-
netic interactions, a genuine revolution has taken place in elementary particle physics in
the last 15 years. One of the basic underlying ideas of these theories is that of sponta-
neous symmetry breaking between different types of interactions due to the appearance
of constant classical scalar fields ϕ over all space (the so-called Higgs fields). Prior to
the appearance of these fields, there is no fundamental difference between strong, weak,
and electromagnetic interactions. Their spontaneous appearance over all space essentially
signifies a restructuring of the vacuum, with certain vector (gauge) fields acquiring high
mass as a result. The interactions mediated by these vector fields then become short-
range, and this leads to symmetry breaking between the various interactions described by
the unified theories.
The first consistent description of strong and weak interactions was obtained within
the scope of gauge theories with spontaneous symmetry breaking. For the first time, it
became possible to investigate strong and weak interaction processes using high-order
perturbation theory. A remarkable property of these theories — asymptotic freedom —
also made it possible in principle to describe interactions of elementary particles up to
center-of-mass energies E ∼ MP ∼ 1019 GeV, that is, up to the Planck energy, where
quantum gravity effects become important.
Here we will recount only the main stages in the development of gauge theories,
rather than discussing their properties in detail. In the 1960s, Glashow, Weinberg, and
Salam proposed a unified theory of the weak and electromagnetic interactions [1], and real
progress was made in this area in 1971–1973 after the theories were shown to be renormal-
izable [2]. It was proved in 1973 that many such theories, with quantum chromodynamics
in particular serving as a description of strong interactions, possess the property of asymp-
totic freedom (a decrease in the coupling constant with increasing energy [3]). The first
unified gauge theories of strong, weak, and electromagnetic interactions with a simple
symmetry group, the so-called grand unified theories [4], were proposed in 1974. The first
theories to unify all of the fundamental interactions, including gravitation, were proposed
in 1976 within the context of supergravity theory. This was followed by the development
of Kaluza–Klein theories, which maintain that our four-dimensional space-time results
from the spontaneous compactification of a higher-dimensional space [6]. Finally, our
most recent hopes for a unified theory of all interactions have been invested in super-
string theory [7]. Modern theories of elementary particles are covered in a number of
 xi

excellent reviews and monographs (see [8–17], for example).

The rapid development of elementary particle theory has not only led to great ad-
vances in our understanding of particle interactions at superhigh energies, but also (as
a consequence) to significant progress in the theory of superdense matter. Only fifteen
years ago, in fact, the term superdense matter meant matter with a density somewhat
higher than nuclear values, ρ ∼ 1014 –1015 g · cm−3 and it was virtually impossible to
conceive of how one might describe matter with ρ ≫ 1015 g · cm−3 . The main problems
15
involved strong-interaction theory, whose typical coupling constants at ρ >∼ 10 g · cm
−3

were large, making standard perturbation-theory predictions of the properties of such

matter unreliable. Because of asymptotic freedom in quantum chromodynamics, how-
ever, the corresponding coupling constants decrease with increasing temperature (and
density). This enables one to describe the behavior of matter at temperatures approach-
ing T ∼ MP ∼ 1019 GeV, which corresponds to a density ρP ∼ M4P ∼ 1094 g · cm−3 .
Present-day elementary particle theories thus make it possible, in principle, to describe
the properties of matter more than 80 orders of magnitude denser than nuclear matter!
The study of the properties of superdense matter described by unified gauge theories
began in 1972 with the work of Kirzhnits [18], who showed that the classical scalar field ϕ
responsible for symmetry breaking should disappear at a high enough temperature T. This
means that a phase transition (or a series of phase transitions) occurs at a sufficiently
high temperature T > Tc , after which symmetry is restored between various types of
interactions. When this happens, elementary particle properties and the laws governing
their interaction change significantly.
This conclusion was confirmed in many subsequent publications [19–24]. It was found
that similar phase transitions could also occur when the density of cold matter was raised
[25–29], and in the presence of external fields and currents [22, 23, 30, 33]. For brevity,
and to conform with current terminology, we will hereafter refer to such processes as phase
transitions in gauge theories.
Such phase transitions typically take place at exceedingly high temperatures and
densities. The critical temperature for a phase transition in the Glashow–Weinberg–
Salam theory of weak and electromagnetic interactions [1], for example, is of the order of
102 GeV ∼ 1015 K. The temperature at which symmetry is restored between the strong
and electroweak interactions in grand unified theories is even higher, Tc ∼ 1015 GeV ∼
1028 K. For comparison, the highest temperature attained in a supernova explosion is
about 1011 K. It is therefore impossible to study such phase transitions in a laboratory.
However, the appropriate extreme conditions could exist at the earliest stages of the
evolution of the universe.
According to the standard version of the hot universe theory, the universe could have
expanded from a state in which its temperature was at least T ∼ 1019 GeV [34, 35],
cooling all the while. This means that in its earliest stages, the symmetry between the
strong, weak, and electromagnetic interactions should have been intact. In cooling, the
universe would have gone through a number of phase transitions, breaking the symmetry
between the different interactions [18–24].
This result comprised the first evidence for the importance of unified theories of ele-
 xii

mentary particles and the theory of superdense matter for the development of the theory
of the evolution of the universe. Cosmologists became particularly interested in recent
theories of elementary particles after it was found that grand unified theories provide a
natural framework within which the observed baryon asymmetry of the universe (that is,
the lack of antimatter in the observable part of the universe) might arise [36–38]. Cos-
mology has likewise turned out to be an important source of information for elementary
particle theory. The recent rapid development of the latter has resulted in a somewhat
unusual situation in that branch of theoretical physics. The reason is that typical el-
ementary particle energies required for a direct test of grand unified theories are of the
order of 1015 GeV, and direct tests of supergravity, Kaluza–Klein theories, and superstring
theory require energies of the order of 1019 GeV. On the other hand, currently planned
accelerators will only produce particle beams with energies of about 104 GeV. Experts
estimate that the largest accelerator that could be built on earth (which has a radius of
about 6000 km) would enable us to study particle interactions at energies of the order
of 107 GeV, which is typically the highest (center-of-mass) energy encountered in cosmic
ray experiments. Yet this is twelve orders of magnitude lower than the Planck energy
EP ∼ MP ∼ 1019 GeV.
The difficulties involved in studying interactions at superhigh energies can be high-
lighted by noting that 1015 GeV is the kinetic energy of a small car, and 1019 GeV is
the kinetic energy of a medium-sized airplane. Estimates indicate that accelerating par-
ticles to energies of the order of 1015 GeV using present-day technology would require an
accelerator approximately one light-year long.
It would be wrong to think, though, that the elementary particle theories currently
being developed are totally without experimental foundation — witness the experiments
on a huge scale which are under way to detect the decay of the proton, as predicted by
grand unified theories. It is also possible that accelerators will enable us to detect some
of the lighter particles (with mass m ∼ 102 –103 GeV) predicted by certain versions of
supergravity and superstring theories. Obtaining information solely in this way, however,
would be similar to trying to discover a unified theory of weak and electromagnetic inter-
actions using only radio telescopes, detecting radio waves with an energy Eγ no greater
EP EW
than 10−5 eV (note that ∼ , where EW ∼ 102 GeV is the characteristic energy in
EW Eγ
the unified theory of weak and electromagnetic interactions).
The only laboratory in which particles with energies of 1015 –1019 GeV could ever exist
and interact with one another is our own universe in the earliest stages of its evolution.
At the beginning of the 1970s, Zeldovich wrote that the universe is the poor man’s
accelerator: experiments don’t need to be funded, and all we have to do is collect the
experimental data and interpret them properly [39]. More recently, it has become quite
clear that the universe is the only accelerator that could ever produce particles at energies
high enough to test unified theories of all fundamental interactions directly, and in that
sense it is not just the poor man’s accelerator but the richest man’s as well. These days,
most new elementary particle theories must first take a “cosmological validity” test —
and only a very few pass.
 xiii

It might seem at first glance that it would be difficult to glean any reasonably definitive
or reliable information from an experiment performed more than ten billion years ago,
but recent studies indicate just the opposite. It has been found, for instance, that phase
transitions, which should occur in a hot universe in accordance with the grand unified
theories, should produce an abundance of magnetic monopoles, the density of which ought
to exceed the observed density of matter at the present time, ρ ∼ 10−29 g · cm−3 , by
approximately fifteen orders of magnitude [40]. At first, it seemed that uncertainties
inherent in both the hot universe theory and the grand unified theories, being very large,
would provide an easy way out of the primordial monopole problem. But many attempts
to resolve this problem within the context of the standard hot universe theory have not
led to final success. A similar situation has arisen in dealing with theories involving
spontaneous breaking of a discrete symmetry (spontaneous CP-invariance breaking, for
example). In such models, phase transitions ought to give rise to supermassive domain
walls, whose existence would sharply conflict with the astrophysical data [41–43]. Going
to more complicated theories such as N = 1 supergravity has engendered new problems
rather than resolving the old ones. Thus it has turned out in most theories based on N = 1
supergravity that the decay of gravitinos (spin = 3/2 superpartners of the graviton) which
existed in the early stages of the universe leads to results differing from the observational
data by about ten orders of magnitude [44, 45]. These theories also predict the existence
of so-called scalar Polonyi fields [15, 46]. The energy density that would have been
accumulated in these fields by now differs from the cosmological data by fifteen orders of
magnitude [47, 48]. A number of axion theories [49] share this difficulty, particularly in
the simplest models based on superstring theory [50]. Most Kaluza–Klein theories based
on supergravity in an 11-dimensional space lead to vacuum energies of order −M4P ∼
−1094 g · cm−3 [16], which differs from the cosmological data by approximately 125 orders
of magnitude. . .
This list could be continued, but as it stands it suffices to illustrate why elementary
particle theorists now find cosmology so interesting and important. An even more gen-
eral reason is that no real unification of all interactions including gravitation is possible
without an analysis of the most important manifestation of that unification, namely the
existence of the universe itself. This is illustrated especially clearly by Kaluza–Klein
and superstring theories, where one must simultaneously investigate the properties of the
space-time formed by compactification of “extra” dimensions, and the phenomenology of
the elementary particles.
It has not yet been possible to overcome some of the problems listed above. This places
important constraints on elementary particle theories currently under development. It is
all the more surprising, then, that many of these problems, together with a number of
others that predate the hot universe theory, have been resolved in the context of one
fairly simple scenario for the development of the universe — the so-called inflationary
universe scenario [51–57]. According to this scenario, the universe, at some very early
stage of its evolution, was in an unstable vacuum-like state and expanded exponentially
(the stage of inflation). The vacuum-like state then decayed, the universe heated up, and
its subsequent evolution can be described by the usual hot universe theory.
 xiv

Since its conception, the inflationary universe scenario has

progressed from something akin to science fiction to a well-established theory of the evo-
lution of the universe accepted by most cosmologists. Of course this doesn’t mean that
we have now finally achieved total enlightenment as to the physical processes operative
in the early universe. The incompleteness of the current picture is reflected by the very
word scenario, which is not normally found in the working vocabulary of a theoretical
physicist. In its present form, this scenario only vaguely resembles the simple models from
which it sprang. Many details of the inflationary universe scenario are changing, tracking
rapidly changing (as noted above) elementary particle theories. Nevertheless, the basic
aspects of this scenario are now well-developed, and it should be possible to provide a
preliminary account of its progress.
Most of the present book is given over to discussion of inflationary cosmology. This
is preceded by an outline of the general theory of spontaneous symmetry breaking and a
discussion of phase transitions in superdense matter, as described by present-day theories
of elementary particles. The choice of material has been dictated by both the author’s
interests and his desire to make the contents useful both to quantum field theorists and
astrophysicists. We have therefore tried to concentrate on those problems that yield an
understanding of the basic aspects of the theory, referring the reader to the original papers
for further details.
In order to make this book as widely accessible as possible, the main exposition has
been preceded by a long introductory chapter, written at a relatively elementary level.
Our hope is that by using this chapter as a guide to the book, and the book itself as a guide
to the original literature, the reader will gradually be able to attain a fairly complete and
accurate understanding of the present status of this branch of science. In this regard, he
might also be assisted by an acquaintance with the books Cosmology of the Early Universe,
by A. D. Dolgov, Ya. B. Zeldovich, and M. V. Sazhin; How the Universe Exploded, by
I. D. Novikov; A Brief History of Time: From the Big Bang to Black Holes, by S. W.
Hawking; and An Introduction to Cosmology and Particle Physics, by R. Dominguez-
Tenreiro and M. Quiros. A good collection of early papers on inflationary cosmology
and galaxy formation can also be found in the book Inflationary Cosmology, edited by L.
Abbott and S.-Y. Pi. We apologize in advance to those authors whose work in the field of
inflationary cosmology we have not been able to treat adequately. Much of the material in
this book is based on the ideas and work of S. Coleman, J. Ellis, A. Guth, S. W. Hawking,
D. A. Kirzhnits, L. A. Kofman, M. A. Markov, V. F. Mukhanov, D. Nanopoulos, I. D.
Novikov, I. L. Rozental’, A. D. Sakharov, A. A. Starobinsky, P. Steinhardt, M. Turner,
and many other scientists whose contribution to modern cosmology could not possibly be
fully reflected in a single monograph, no matter how detailed.
I would like to dedicate this book to the memory of Yakov Borisovich Zeldovich, who
should by rights be considered the founder of the Soviet school of cosmology.
 1
Overview of Unified Theories of Elementary
Particles and the Inflationary Universe
Scenario

1.1 The scalar field and spontaneous symmetry breaking

Scalar fields ϕ play a fundamental role in unified theories of the weak, strong, and elec-
tromagnetic interactions. Mathematically, the theory of these fields is simpler than that
of the spinor fields ψ describing electrons or quarks, for instance, and it is simpler than
the theory of the vector fields Aµ which describes photons, gluons, and so on. The most
interesting and important properties of these fields for both elementary particle theory
and cosmology, however, were grasped only fairly recently.
Let us recall the basic properties of such fields. Consider first the simplest theory of
a one-component real scalar field ϕ with the Lagrangian1

1 m2 2 λ 4
L= (∂µ ϕ)2 − ϕ − ϕ . (1.1.1)
2 2 4
In this equation, m is the mass of the scalar field, and λ is its coupling constant. For
simplicity, we assume throughout that λ ≪ 1. When ϕ is small and we can neglect the
last term in (1.1.1), the field satisfies the Klein–Gordon equation

( + m2 ) ϕ = ϕ̈ − ∆ϕ + m2 ϕ = 0 , (1.1.2)

where a dot denotes differentiation with respect to time. The general solution of this
equation is expressible as a superposition of plane waves, corresponding to the propagation
1
In this book we employ units such that h̄ = c = 1, the system commonly used in elementary particle
theory. In order to transform expressions to conventional units, corresponding terms must be multiplied
by appropriate powers of h̄ or c to give the correct dimensionality (note that h̄ = 6.6 · 10−22 MeV · sec ≈
10−27 erg · sec, c ≈ 3 · 1010 cm · sec−1 ). Thus, for example Eq. (1.1.1) would acquire the form

1 m 2 c2 2 λ 4
L= (∂µ ϕ)2 − ϕ − ϕ .
2 2h̄2 4
 2

V V

0 ϕ 0 ϕ0 ϕ
a b
Figure 1.1: Effective potential V(ϕ) in the simplest theories of the scalar field ϕ. a) V(ϕ)
in the theory (1.1.1), and b) in the theory (1.1.5).

of particles of mass m and momentum k [58]:

Z
ϕ(x) = (2π)−3/2 d4 k δ(k 2 − m2 )[ei k x ϕ+ (k) + e−i k x ϕ− (k)]
d3 k
Z
= (2π) −3/2
√ [ei k x a+ (k) + e−i k x a− (k)] , (1.1.3)
2k0
1 √
where a± (k) = √ ϕ± (k), k0 = k2 + m2 , k x = k0 t − k · x. According to (1.1.3),
2k0
the field ϕ(x) will oscillate about the point ϕ = 0 density for the field ϕ (the so-called
effective potential)
1 m2 2 λ 4
V(ϕ) = (∇ϕ)2 + ϕ + ϕ (1.1.4)
2 2 4
occurs at ϕ = 0 (see Fig. 1.1a).
Fundamental advances in the unification of the weak, strong, and electromagnetic
interactions were finally achieved when simple theories based on Lagrangians like (1.1.1)
with m2 > 0 gave way to what were at first glance somewhat strange-looking theories
with negative mass squared:
1 µ2 2 λ 4
L= (∂µ ϕ)2 + ϕ − ϕ . (1.1.5)
2 2 4
Instead of oscillations about ϕ = 0, the solution corresponding to (1.1.3) gives modes
that grow exponentially near ϕ = 0 when k2 < m2 :
 q 
δϕ(k) ∼ exp ± µ2 − k2 t · exp(±i k x) . (1.1.6)

What this means is that the minimum of the effective potential

1 µ2 2 λ 4
V(ϕ) = (∇ϕ)2 − ϕ + ϕ (1.1.7)
2 2 4
 3
√
will now occur not at ϕ = 0, but at ϕc = ±µ/ λ (see Fig. 1.1b).2 Thus, even if the
field ϕ is zero initially, it soon undergoes a transition
√ (after a time of order µ−1 ) to a
stable state with the classical field ϕc = ±µ/ λ, a phenomenon known as spontaneous
symmetry breaking. √
After spontaneous symmetry breaking, excitations of the field ϕ near ϕc = ±µ/ λ
can also be described by a solution like (1.1.3). In order to do so, we make the change of
variables
ϕ → ϕ + ϕ0 . (1.1.8)
The Lagrangian (1.1.5) thereupon takes the form
1 µ2 λ
L(ϕ + ϕ0 ) = (∂µ (ϕ + ϕ0 ))2 + (ϕ + ϕ0 )2 − (ϕ + ϕ0 )4
2 2 4
2 2
1 3 λ ϕ 0 − µ λ
= (∂µ ϕ)2 − ϕ2 − λ ϕ0 ϕ3 − ϕ4
2 2 4
2
µ 2 λ 4
+ ϕ0 − ϕ0 − ϕ (λϕ20 − µ2 ) ϕ0 . (1.1.9)
2 4
We see from (1.1.9) that when ϕ0 =
6 0, the effective mass squared of the field ϕ is not
equal to −µ2 , but rather
m2 = 3 λ ϕ20 − µ2 , (1.1.10)
√
and when ϕ0 = ±µ/ λ, at the minimum of the potential V(ϕ) given by (1.1.7), we have

m2 = 2 λ ϕ20 = 2 µ2 > 0 ; (1.1.11)

in other words, the mass squared of the field ϕ has the correct sign. Reverting to the
original variables, we can write the solution for ϕ in the form
d3 k i k x +
Z
ϕ(x) = ϕ0 + (2π) −3/2
√ [e a (k) + e−i k x a− (k)] . (1.1.12)
2k0
The integral in (1.1.12) corresponds to particles (quanta) of the field ϕ with mass given
by (1.1.11), propagating against the background of the constant classical field ϕ0 .
The presence of the constant classical field ϕ0 over all space will not give rise to any
preferred reference frame associated with that field: the Lagrangian (1.1.9) is covariant,
irrespective of the magnitude of ϕ0 . Essentially, the appearance of a uniform field ϕ0 over
all space simply represents a restructuring of the vacuum state. In that sense, the space
filled by the field ϕ0 remains “empty.” Why then is it necessary to spoil the good theory
(1.1.1)?
The main point here is that the advent of the field ϕ0 changes the masses of those
particles with which it interacts. We have already seen this in considering the example of
the sign “correction” for the mass squared of the field ϕ in the theory (1.1.5). Similarly,
scalar fields can change the mass of both fermions and vector particles.
2
V(ϕ) usually attains a minimum for homogeneous fields ϕ, so gradient terms in the expression for V(ϕ)
are often omitted.
 4

Let us examine the two simplest models. The first is the simplified σ-model, which is
sometimes used for a phenomenological description of strong interactions at high energy
[26]. The Lagrangian for this model is a sum of the Lagrangian (1.1.5) and the Lagrangian
for the massless fermions ψ, which interact with ϕ with a coupling constant h:
1 µ2 2 λ 4
L= (∂µ ϕ)2 + ϕ − ϕ + ψ̄ (i ∂µ γµ − h ϕ) ψ . (1.1.13)
2 2 4
After symmetry breaking, the fermions will clearly acquire a mass
µ
mψ = h |ϕ0 | = h √ . (1.1.14)
λ
The second is the so-called Higgs model [59], which describes an Abelian vector field
Aµ (the analog of√ the electromagnetic field) that interacts with the complex scalar field
χ = (χ1 + i χ2 )/ 2. The Lagrangian for this theory is given by
1
L = − (∂µ Aν − ∂ν Aµ )2 + (∂µ + i e Aµ ) χ∗ (∂µ − i e Aµ ) χ
4
+ µ χ χ − λ (χ∗ χ)2 .
2 ∗
(1.1.15)
As in (1.1.7), when µ2 < 0 the scalar field χ acquires a classical component. This effect
is described most easily by making the change of variables
1 i ζ(x)
χ(x) → √ (ϕ(x) + ϕ0 ) exp ,
2 ϕ0
1
Aµ (x) → Aµ (x) + ∂µ ζ(x) , (1.1.16)
e ϕ0
whereupon the Lagrangian (1.1.15) becomes
1 e2 1
L = − (∂µ Aν − ∂ν Aµ ) + (ϕ + ϕ0 )2 A2µ + (∂µ ϕ)2
2
4 2 2
3 λ ϕ20 − µ2 2 λ µ 2
λ
− ϕ − λ ϕ0 ϕ3 − ϕ4 + ϕ20 − ϕ40
2 4 2 4
2 2
− ϕ(λ ϕ0 − µ ) ϕ0 . (1.1.17)
Notice that the auxiliary field ζ(x) has been entirely cancelled out of (1.1.17), which
describes a theory of vector particles of mass mA = e ϕ0 that interact with a scalar field
2
having the effective
√ potential (1.1.7). As before, when µ > 0, symmetry breaking occurs,
√
the field ϕ0 = µ/ λ appears, and the vector particles of Aµ acquire a mass mA = e µ/ λ.
This scheme for making vector mesons massive is called the Higgs mechanism, and the
fields χ, ϕ are known as Higgs fields. The appearance of the classical field ϕ0 breaks the
symmetry of (1.1.15) under U(1) gauge transformations:
1
Aµ → Aµ + ∂µ ζ(x)
e
χ → χ exp [i ζ(x)] . (1.1.18)
 5

The basic idea underlying unified theories of the weak, strong, and electromagnetic
interactions is that prior to symmetry breaking, all vector mesons (which mediate these in-
teractions) are massless, and there are no fundamental differences among the interactions.
As a result of the symmetry breaking, however, some of the vector bosons do acquire mass,
and their corresponding interactions become short-range, thereby destroying the symme-
try between the various interactions. For example, prior to the appearance of the constant
scalar Higgs field H, the Glashow–Weinberg–Salam model [1] has SU(2) × U(1) symmetry,
and electroweak interactions are mediated by massless vector bosons. After the appear-
ance of the constant scalar field H, some of the vector bosons (Wµ± and Z0µ ) acquire masses
of order eH ∼ 100 GeV, and the corresponding interactions become short-range (weak
interactions), whereas the electromagnetic field Aµ remains massless.
The Glashow–Weinberg–Salam model was proposed in the 1960’s [1], but the real ex-
plosion of interest in such theories did not come until 1971–1973, when it was shown that
gauge theories with spontaneous symmetry breaking are renormalizable, which means that
there is a regular method for dealing with the ultraviolet divergences, as in ordinary quan-
tum electrodynamics [2]. The proof of renormalizability for unified field theories is rather
complicated, but the basic physical idea behind it is quite simple. Before the appearance
of the scalar field ϕ0 , the unified theories are renormalizable, just like ordinary quantum
electrodynamics. Naturally, the appearance of a classical scalar field ϕ0 (like the presence
of the ordinary classical electric and magnetic fields) should not affect the high-energy
properties of the theory; specifically, it should not destroy the original renormalizability of
the theory. The creation of unified gauge theories with spontaneous symmetry breaking
and the proof that they are renormalizable carried elementary particle theory in the early
1970’s to a qualitatively new level of development.
The number of scalar field types occurring in unified theories can be quite large. For
example, there are two Higgs fields in the simplest theory with SU(5) symmetry [4]. One
of these, the field Φ, is represented by a traceless 5 × 5 matrix. Symmetry breaking in
this theory results from the appearance of the classical field
 
1 0
1
s  
2
 
 
Φ0 = ϕ0  1  , (1.1.19)
15  

 −3/2 

0 −3/2
where the value of the field ϕ0 is extremely large — ϕ0 ∼ 1015 GeV. All vector particles
in this theory are massless prior to symmetry breaking, and there is no fundamental
difference between the weak, strong, and electromagnetic interactions. Leptons can then
easily be transformed into quarks, and vice versa. After the appearance of the field
(1.1.19), some of the vector mesons (the X and Y mesons responsible for transforming
quarks into leptons) acquire enormous mass: mX,Y = (5/3)1/2 g ϕ0 /2 ∼ 1015 GeV, where
g 2 ∼ 0.3 is the SU(5) gauge coupling constant. The transformation of quarks into leptons
thereupon becomes strongly inhibited, and the proton becomes almost stable. The original
SU(5) symmetry breaks down into SU(3) × SU(2) × U(1); that is, the strong interactions
 6

(SU(3)) are separated from the electroweak (SU(2) × U(1)). Yet another classical scalar
field H ∼ 102 GeV then makes its appearance, breaking the symmetry between the weak
and electromagnetic interactions, as in the Glashow–Weinberg–Salam theory [4, 12].
The Higgs effect and the general properties of theories with spontaneous symmetry
breaking are discussed in more detail in Chapter 2. The elementary theory of sponta-
neous symmetry breaking is discussed in Section 2.1. In Section 2.2, we further study
this phenomenon, with quantum corrections to the effective potential V(ϕ) taken into
consideration. As will be shown in Section 2.2, quantum corrections can in some cases
significantly modify the general form of the potential (1.1.7). Especially interesting and
unexpected properties of that potential will become apparent when we study it in the
1/N approximation.

1.2 Phase transitions in gauge theories

The idea of spontaneous symmetry breaking, which has proven so useful in building unified
gauge theories, has an extensive history in solid-state theory and quantum statistics,
where it has been used to describe such phenomena as ferromagnetism, superfluidity,
superconductivity, and so forth.
Consider, for example, the expression for the energy of a superconductor in the phe-
nomenological Ginzburg–Landau theory [60] of superconductivity:
H2 1
E = E0 + + |(∇ − 2 i e A) Ψ|2 − α |Ψ|2 + β |Ψ|4 . (1.2.1)
2 2m
Here E0 is the energy of the normal metal without a magnetic field H, Ψ is the field
describing the Cooper-pair Bose condensate, and α and β are positive parameters.
Bearing in mind, then, that the potential energy of a field enters into the Lagrangian
with a negative sign, it is not hard to show that the Higgs model (1.1.15) is simply a rel-
ativistic generalization of the Ginzburg–Landau theory of superconductivity (1.2.1), and
the classical field ϕ in the Higgs model is the analog of the Cooper-pair Bose condensate.3
The analogy between unified theories with spontaneous symmetry breaking and theo-
ries of superconductivity has been found to be extremely useful in studying the properties
of superdense matter described by unified theories. Specifically, it is well known that when
the temperature is raised, the Cooper-pair condensate shrinks to zero and superconduc-
tivity disappears. It turns out that the uniform scalar field ϕ should also disappear when
the temperature of matter is raised; in other words, at superhigh temperatures, the sym-
metry between the weak, strong, and electromagnetic interactions ought to be restored
[18–24].
A theory of phase transitions involving the disappearance of the classical field ϕ is
discussed in detail in Ref. 24. In gross outline, the basic idea is that the equilibrium
3
Where this does not lead to confusion, we will simply denote the classical scalar field by ϕ, rather then
ϕ0 . In certain other cases, we will also denote the initial value of the classical scalar field ϕ by ϕ0 . We
hope that the meaning of ϕ and ϕ0 in each particular case will be clear from the context.
 7

V(ϕ) – V(0)
C B A

0 ϕ

Figure 1.2: Effective potential V(ϕ, T) in the theory (1.1.5) at finite temperature. A)
T = 0; B) 0 < T < Tc ; C) T > Tc . As the temperature rises, the field ϕ varies smoothly,
corresponding to a second-order phase transition.

value of the field ϕ at fixed temperature T 6= 0 is governed not by the location of the
minimum of the potential energy density V(ϕ), but by the location of the minimum of
the free energy density F(ϕ, T) ≡ V(ϕ, T), which equals V(ϕ) at T = 0. It is well-known
that the temperature-dependent contribution to the free energy F from ultrarelativistic
scalar particles of mass m at temperature T ≫ m is given [61] by
π2 m2 2 m
  
∆F = ∆V(ϕ, T) = − T4 + T 1+O . (1.2.2)
90 24 T
If we then recall that
d2 V
m2 (ϕ) = = 3 λ ϕ2 − µ 2
dϕ2
in the model (1.1.5) (see Eq. (1.1.10)), the complete expression for V(ϕ, T) can be written
in the form
µ2 λ ϕ4 λ T2 2
V(ϕ, T) = − ϕ2 + + ϕ + ... , (1.2.3)
2 4 8
where we have omitted terms that do not depend on ϕ. The behavior of V(ϕ, T) is shown
in Fig. 1.2 for a number of different temperatures.
It is clear from (1.2.3) that as T rises, the equilibrium value of ϕ at the minimum of
V(ϕ, T) decreases, and above some critical temperature
2µ
Tc = √ , (1.2.4)
λ
the only remaining minimum is the one at ϕ = 0, i.e., symmetry is restored (see Fig. 1.2).
Equation (1.2.3) then implies that the field ϕ decreases continuously to zero with rising
temperature; the restoration of symmetry in the theory (1.1.5) is a second-order phase
transition.
 8

V(ϕ) – V(0)
D C B A

0 ϕ

Figure 1.3: Behavior of the effective potential V(ϕ, T) in theories in which phase tran-
sitions are first-order. Between Tc1 and Tc2 , the effective potential has two minima;
at T = Tc , these minima have the same depth. A) T = 0; B) Tc1 < T < Tc ; C)
Tc < T < Tc2 ; D) T > Tc2 .

Note that in the case at hand, when λ ≪ 1, Tc ≫ m over the entire range of values
of ϕ that is of interest (ϕ <∼ ϕc ), so that a high-temperature expansion of V(ϕ, T) in
powers of m/T in (1.2.2) is perfectly justified. However, it is by no means true that
phase transitions take place only at T ≫ m in all theories. It often happens that at the
instant of a phase transition, the potential V(ϕ, T) has two local minima, one giving a
stable state and the other an unstable state of the system (Fig. 1.3). We then have a
first-order phase transition, due to the formation and subsequent expansion of bubbles
of a stable phase within an unstable one, as in boiling water. Investigation of the first-
order phase transitions in gauge theories [62] indicates that such transitions are sometimes
considerably delayed, so that the transition takes place (with rising temperature) from a
strongly superheated state, or (with falling temperature) from a strongly supercooled one.
Such processes are explosive, which can lead to many important and interesting effects
in an expanding universe. The formation of bubbles of a new phase is typically a barrier
tunnelling process; the theory of this process at a finite temperature was given in [62].
It is well known that superconductivity can be destroyed not only by heating, but also
by external fields H and currents j; analogous effects exist in unified gauge theories [22,
23]. On the other hand, the value of the field ϕ, being a scalar, should depend not just
on the currents j, but on the square of current j 2 = ρ2 − j2 , where ρ is the charge density.
Therefore, while increasing the current j usually leads to the restoration of symmetry
in gauge theories, increasing the charge density ρ usually results in the enhancement of
symmetry breaking [27]. This effect and others that may exist in superdense cold matter
are discussed in Refs. 27–29.
 9

1.3 Hot universe theory

There have been two important stages in the development of twentieth-century cosmology.
The first began in the 1920’s, when Friedmann used the general theory of relativity to
create a theory of a homogeneous and isotropic expanding universe with metric [63–65]

dr 2
" #
2 2 2
ds = dt − a (t) + r 2 (dθ + sin2 θ dϕ2 ) , (1.3.1)
1 − k r2
where k = +1, −1, or 0 for a closed, open, or flat Friedmann universe, and a(t) is the
“radius” of the universe, or more precisely, its scale factor (the total size of the universe
may be infinite). The term flat universe refers to the fact that when k = 0, the metric
(1.3.1) can be put in the form

ds2 = dt2 − a2 (t) (dx2 + dy 2 + dz 2 ) . (1.3.2)

At any given moment, the spatial part of the metric describes an ordinary three-dimensional
Euclidean (flat) space, and when a(t) is constant (or slowly varying, as in our universe at
present), the flat-universe metric describes Minkowski space.
For k = ±1, the geometrical interpretation of the three-dimensional space part of
(1.3.1) is somewhat more complicated [65]. The analog of a closed world at any given time t
is a sphere S3 embedded in some auxiliary four-dimensional space (x, y, z, τ ). Coordinates
on this sphere are related by

x2 + y 2 + z 2 + τ 2 = a2 (t) . (1.3.3)

The metric on the surface can be written in the form

dr 2
" #
2 2
dl = a (t) + r 2 (dθ2 + sin2 θ dϕ2 ) , (1.3.4)
1 − r2

where r, θ, and ϕ are spherical coordinates on the surface of the sphere S3 .

The analog of an open universe at fixed t is the surface of the hyperboloid

x2 + y 2 + z 2 − τ 2 = a2 (t) . (1.3.5)

The evolution of the scale factor a(t) is given by the Einstein equations
4π
ä = − G (ρ + 3 p) a , (1.3.6)
3
 2
k ȧ k 8π
H2 + ≡ + 2 = Gρ . (1.3.7)
a2 a a 3
Here ρ is the energy density of matter in the universe, and p is its pressure. The gravi-
ȧ
tational constant G = M−2
P , where MP = 1.2 · 10
19
GeV is the Planck mass,4 and H =
a
4
The reader should be warned that in the recent literature the authors often use√a different definition of
the Planck mass, which is smaller than the one used in our book by a factor of 8π.
 10

a
O
F

C
?
0 tc t

Figure 1.4: Evolution of the scale factor a(t) for three different versions of the Friedmann
hot universe theory: open (O), flat (F), and closed (C).

is the Hubble “constant”, which in general is a function of time. Equations (1.3.6) and
(1.3.7) imply an energy conservation law, which can be written in the form

ρ̇ a3 + 3 (ρ + p) a2 ȧ = 0 . (1.3.8)

To find out how this universe will evolve in time, one also needs to know the so-called
equation of state, which relates the energy density of matter to its pressure. One may
assume, for instance, that the equation of state for matter in the universe takes the form
p = α ρ. From the energy conservation law, one then deduces that

ρ ∼ a−3(1+α) . (1.3.9)

In particular, for nonrelativistic cold matter with p = 0,

ρ ∼ a−3 , (1.3.10)
ρ
and for a hot ultrarelativistic gas of noninteracting particles with p = ,
3
ρ ∼ a−4 . (1.3.11)
ρ
In either case (and in general for any medium with p > − ), when a is small, the quantity
3
8π k
G ρ is much greater than 2 . We then find from (1.3.7) that for small a, the expansion
3 a
of the universe goes as
2
a ∼ t 3(1+a) . (1.3.12)
In particular, for nonrelativistic cold matter

a ∼ t2/3 , (1.3.13)

and for the ultrarelativistic gas

a ∼ t1/2 . (1.3.14)
 11

Thus, regardless of the model used (k = ±1, 0), the scale factor vanishes at some time
t = 0, and the matter density at that time becomes infinite. It can also be shown that
at that time, the curvature tensor Rµναβ goes to infinity as well. That is why the point
t = 0 is known as the point of the initial cosmological singularity (Big Bang).
An open or flat universe will continue to expand forever. In a closed universe with
ρ
p > − , on the other hand, there will be some point in the expansion when the term
3
1 8π
2
in (1.3.7) becomes equal to G ρ. Thereafter, the scale constant a decreases, and it
a 3
vanishes at some time tc (Big Crunch). It is straightforward to show [65] that the lifetime
of a closed universe filled with a total mass M of cold nonrelativistic matter is
4M 4M M
tc = G= 2
∼ · 10−43 sec . (1.3.15)
3 3 MP MP

The lifetime of a closed universe filled with a hot ultrarelativistic gas of particles of a
single species may be conveniently expressed in terms of the total entropy of the universe,
S = 2 π 2 a3 s, where s is the entropy density. If the total entropy of the universe does not
change (adiabatic expansion), as is often assumed, then
1/6
32 S2/3

tc = ∼ S2/3 · 10−43 sec . (1.3.16)
45 π 2 MP
These estimates will turn out to be useful in discussing the difficulties encountered by the
standard theory of expansion of the hot universe.
Up to the mid-1960’s, it was still not clear whether the early universe had been hot or
cold. The critical juncture marking the beginning of the second stage in the development
of modern cosmology was Penzias and Wilson’s 1964–65 discovery of the 2.7 K microwave
background radiation arriving from the farthest reaches of the universe. The existence of
the microwave background had been predicted by the hot universe theory [66, 67], which
gained immediate and widespread acceptance after the discovery.
According to that theory, the universe, in the very early stages of its evolution, was
filled with an ultrarelativistic gas of photons, electrons, positrons, quarks, antiquarks,
etc. At that epoch, the excess of baryons over antibaryons was but a small fraction (at
most 10−9 ) of the total number of particles. As a result of the decrease of the effective
coupling constants for weak, strong, and electromagnetic interactions with increasing
density, effects related to interactions among those particles affected the equation of state
of the superdense matter only slightly, and the quantities s, ρ, and p were given [61] by

π2
ρ = 3p = N(T) T4 , (1.3.17)
30
2 π2
s = N(T) T3 , (1.3.18)
45
7
where the effective number of particle species N(T) is NB (T) + NF (T), and NB and NF
8
 12

are the number of boson and fermion species5 with masses m ≪ T.

In realistic elementary particle theories, N(T) increases with increasing T, but it typi-
cally does so relatively slowly, varying over the range 102 to 104 . If the universe expanded
adiabatically, with s a3 ≈ const, then (1.3.18) implies that during the expansion, the
quantity aT also remained approximately constant. In other words, the temperature of
the universe dropped off as
T(t) ∼ a−1 (t) . (1.3.19)
The background radiation detected by Penzias and Wilson is a result of the cooling
of the hot photon gas during the expansion of the universe. The exact equation for the
time-dependence of the temperature in the early universe can be derived from (1.3.7) and
(1.3.17): s
1 45 MP
t= . (1.3.20)
4 π π N(T) T2
In the later stages of the evolution of the universe, particles and antiparticles annihilate
each other, the photon-gas energy density falls off relatively rapidly (compare (1.3.10)
and (1.3.11)), and the main contribution to the matter density starts to come from the
small excess of baryons over antibaryons, as well as from other fields and particles which
now comprise the so-called hidden mass in the universe.
The most detailed and accurate description of the hot universe theory can be found
in the fundamental monograph by Zeldovich and Novikov [34] (see also [35]).
Several different avenues were pursued in the 1970’s in developing this theory. Two
of these will be most important in the subsequent discussion: the development of the
hot universe theory with regard to the theory of phase transitions in superdense matter
[18–24], and the theory of formation of the baryon asymmetry of the universe [36–38].
Specifically, as just stated in the preceding paragraph, symmetry should be restored
in grand unified theories at superhigh temperatures. As applied to the simplest SU(5)
15
model, for instance, this means that at a temperature T >∼ 10 GeV, there was essentially
no difference between the weak, strong, and electromagnetic interactions, and quarks
could easily transform into leptons; that is, there was no such thing as baryon number
conservation. At t1 ∼ 10−35 sec after the Big Bang, when the temperature had dropped
to T ∼ Tc1 ∼ 1014 –1015 GeV, the universe underwent the first symmetry-breaking phase
transition, with SU(5) perhaps being broken into SU(3) × SU(2) × U(1). After this
transition, strong interactions were separated from electroweak and leptons from quarks,
and superheavy-meson decay processes ultimately leading to the baryon asymmetry of the
universe were initiated. Then, at t2 ∼ 10−10 sec, when the temperature had dropped to
Tc2 ∼ 102 GeV, there was a second phase transition, which broke the symmetry between
the weak and electromagnetic interactions, SU(3) × SU(2) × U(1) → SU(3) × U(1). As
the temperature dropped still further to Tc3 ∼ 102 MeV, there was yet another phase
transition (or perhaps two distinct ones), with the formation of baryons and mesons
from quarks and the breaking of chiral invariance in strong interaction theory. Physical
5
To be more precise, NB and NF are the number of boson and fermion degrees of freedom. For example,
NB = 2 for photons, NF = 2 for neutrinos, NF = 4 for electrons, etc.
 13

processes taking place at later stages in the evolution of the universe were much less
dependent on the specific features of unified gauge theories (a description of these processes
can be found in the books cited above [34, 35]).
Most of what we have to say in this book will deal with events that transpired approx-
imately 1010 years ago, in the time up to about 10−10 seconds after the Big Bang. This
will make it possible to examine the global structure of the universe, to derive a more
adequate understanding of the present state of the universe and its future, and finally,
even to modify considerably the very notion of the Big Bang.

1.4 Some properties of the Friedmann models

In order to provide some orientation for the problems of modern cosmology, it is neces-
sary to present at least a rough idea of typical values of the quantities appearing in the
equations, the relationships among these quantities, and their physical meaning.
We start with the Einstein equation (1.3.7), which we will find to be particularly
ȧ
important in what follows. What can one say about the Hubble parameter H = , the
a
density ρ, and the quantity k?
At the earliest stages of the evolution of the universe (not long after the singularity),
H and ρ might have been arbitrarily large. It is usually assumed, though, that at densities
4 94 3
ρ>∼ MP ∼ 10 g/cm , quantum gravity effects are so significant that quantum fluctuations
of the metric exceed the classical value of gµν , and classical space-time does not provide
an adequate description of the universe [34]. We therefore restrict further discussion
4 19
to phenomena for which ρ < ∼ MP , T <∼ MP ∼ 10 GeV, H < MP , and so on. This
restriction can easily be made more precise by noting that quantum corrections to the
MP
Einstein equations in a hot universe are already significant for T ∼ √ ∼ 1017 –1018
N
M4P
GeV and ρ ∼ ∼ 1090 –1092 g/cm3 . It is also worth noting that in an expanding
N
universe, thermodynamic equilibrium cannot be established immediately, but only when
the temperature T is sufficiently low. Thus in SU(5) models, for example, the typical
time for equilibrium to be established is only comparable to the age t of the universe from
16
(1.3.20) when T < ∗
∼ T ∼ 10 GeV (ignoring hypothetical graviton processes that might
lead to equilibrium even before the Planck time has elapsed, with ρ ≫ M4P ).
The behavior of the nonequilibrium universe at densities of the order of the Planck
density is an important problem to which we shall return again and again. Notice, how-
ever, that T∗ ∼ 1016 GeV exceeds the typical critical temperature for a phase transition
15
in grand unified theories, Tc <∼ 10 GeV.
At the present time, the values of H and ρ are not well-determined. For example,
km
H = 100 h ∼ h · (3 · 1017 )−1 sec−1 ∼ h · 10−10 yr−1 , (1.4.1)
sec · Mpc
 14

where the factor h = 0.7±0.03 (1 megaparsec (Mpc) equals 3.09·1024 cm or 3.26·106 light
years). For a flat universe, H and ρ are uniquely related by Eq. (1.3.7); the corresponding
value ρ = ρc (H) is known as the critical density, since the universe must be closed (for
given H) at higher density, and open at lower:

3 H2 3 H2 M2P
ρc = = , (1.4.2)
8πG 8π
and at present, the critical density of the universe is

ρc ≈ 2 · 10−29 h2 g/cm3 . (1.4.3)

The ratio of the actual density of the universe to the critical density is given by the
quantity Ω,
ρ
Ω= . (1.4.4)
ρc
Contributions to the density ρ come both from luminous baryon matter, with ρLB ∼
10−2 ρc , and from dark (hidden, missing) matter, which should have a density at least an
order of magnitude higher. The observational data imply that6

Ω = 1.01 ± 0.02. (1.4.5)

The present-day universe is thus not too far from being flat (while according to the
inflationary universe scenario, Ω = 1 to high accuracy; see below). Furthermore, as we
remarked previously, the early universe not far from being spatially flat because of the
k 8πG
relatively small value of 2 compared to ρ in (1.3.7). From here on, therefore, we
a 3
confine our estimates to those for a flat universe (k = 0).
Equations (1.3.13) and (1.3.14) imply that the age of a universe filled with ultrarela-
ȧ
tivistic gas is related to the quantity H = by
a
1
t= , (1.4.6)
2H
and for a universe with the equation of state p = 0,
2
t= . (1.4.7)
3H
If, as is often supposed, the major contribution to the missing mass comes from nonrela-
tivistic matter, the age of the universe will presently be given by Eq. (1.4.7):
2 1
t∼ · 1010 yr ; <h<1. (1.4.8)
3h 2 ∼ ∼
6
The estimate of h and Ω are changed from their values given in the original edition of the book with an
account taken of the recent observational data.
 15

H(t) not only determines the age, but the distance to the horizon as well, that is, the
radius of the observable part of the universe.
To be more precise, one must distinguish between two horizons — the particle horizon
and the event horizon [35].
The particle horizon delimits the causally connected part of the universe that an
observer can see in principle at a given time t. Since light propagates on the light cone
ds2 = 0, we find from (1.3.1) that the rate at which the radius r of a wavefront changes is
√
dr 1 − k r2
= , (1.4.9)
dt a(t)

and the physical distance travelled by light in time t is

r(t) dr t dt′
Z Z
Rp (t) = a(t) √ = a(t) . (1.4.10)
0 1 − k r2 0 a(t′ )

In particular, for a(t) ∼ t3/2 (1.3.13),

Rp = 3 t = 2 [H(t)]−1 . (1.4.11)

The quantity Rp gives the size of the observable part of the universe at time t. From
(1.4.1) and (1.4.11), we obtain the present-day value of Rp (i.e., the distance to the particle
horizon) for the cold dark matter dominated universe

Rp ∼ 2 h−1 · 1028 cm . (1.4.12)

In a certain conceptual sense, the event horizon is the complement of the particle
horizon: it delimits that part of the universe from which we can ever (up to some time
tmax ) receive information about events taking place now (at time t):
tmax dt′
Z
Re (t) = a(t) . (1.4.13)
t a(t′ )

For a flat universe with a(t) ∼ t2/3 , there is no event horizon: Re (t) → ∞ as tmax → ∞.
In what follows, we will be particularly interested in the case a(t) ∼ eHt , where H = const.
This corresponds to the Sitter metric, and gives

Re (t) = H−1 . (1.4.14)

The thrust of this result is that an observer in an exponentially expanding universe sees
only those events that take place at a distance no farther away than H−1 . This is com-
pletely analogous to the situation for a black hole, from whose surface no information
can escape. The difference is that an observer in de Sitter space (in an exponentially
expanding universe) will find himself effectively surrounded by a “black hole” located at
a distance H−1 .
 16

In closing, let us note one more rather perplexing circumstance. Consider two points
separated by a distance R at time t in a flat Friedmann universe. If the spatial coordinates
of these points remain unchanged (and in that sense, they remain stationary), the distance
between them will nevertheless increase, due to the general expansion of the universe, at
a rate
dR ȧ
= R = HR . (1.4.15)
dt a
What this means, then, is that two points more than a distance H−1 apart will move away
from one another faster than the speed of light c = 1. But there is no paradox here, since
what we are concerned with now is the rate at which two objects subject to the general
cosmological expansion separate from each other, and not with a signal propagation ve-
locity at all, which is related to the local variation of particle spatial coordinates. On the
other hand, it is just this effect that provides the foundation for the existence of an event
horizon in de Sitter space.

1.5 Problems of the standard scenario

Following the discovery of the microwave background radiation, the hot universe theory
immediately gained widespread acceptance. Workers in the field have indeed pointed
out certain difficulties which, over the course of many years, have nevertheless come to
be looked upon as only temporary. In order to make the changes now taking place in
cosmology more comprehensible, we list here some of the problems of the standard hot
universe theory.

1.5.1. The singularity problem

Equations (1.3.9) and (1.3.12) imply that for all “reasonable” equations of state, the
density of matter in the universe goes to infinity as t → 0, and the corresponding solutions
cannot be formally continued to the domain t < 0.
One of the most distressing questions facing cosmologists is whether anything existed
before t = 0; if not, then where did the universe come from? The birth and death of
the universe, like the birth and death of a human being, is one of the most worrisome
problems facing not just cosmologists, but all of contemporary science.
At first, there seemed to be some hope that even if the problem could not be solved,
it might at least be possible to circumvent it by considering a more general model of the
universe than the Friedmann model — perhaps an inhomogeneous, anisotropic universe
filled with matter having some exotic equation of state. Studies of the general structure of
space-time near a singularity [68] and several important theorems on singularities in the
general theory of relativity [69, 70] proven by topological methods, however, demonstrated
that it was highly unlikely that this problem could be solved within the framework of
classical gravitation theory.
 17

1.5.2. The flatness of space

This problem admits of several equivalent or almost equivalent formulations, differing

somewhat in the approach taken.

a. THE EUCLIDICITY PROBLEM. We all learned in grade school that our world
is described by Euclidean geometry, in which the angles of a triangle sum to 180◦ and
parallel lines never meet (or they “meet at infinity”). In college, we were told that it
was Riemannian geometry that described the world, and that parallel lines could meet or
diverge at infinity. But nobody ever explained why what we learned in school was also
true (or almost true) — that is, why the world is Euclidean to such an incredible degree
of accuracy. This is even more surprising when one realizes that there is but one natural
scale length in general relativity, the Planck length lP ∼ M−1 P ∼ 10
−33
cm.
One might expect that the world would be close to Euclidean except perhaps at
distances of the order of lP or less (that is, less than the characteristic radius of curvature
of space). In fact, the opposite is true: on small scales l < ∼ lP , quantum fluctuations of
the metric make it impossible in general to describe space in classical terms (this leads
to the concept of space-time foam [71]). At the same time, for reasons unknown, space is
almost perfectly Euclidean on large scales, up to l ∼ 1028 cm — 60 orders of magnitude
greater than the Planck length.

b. THE FLATNESS PROBLEM. The seriousness of the preceding problem is most

easily appreciated in the context of the Friedmann model (1.3.1). We have from Eq.
(1.3.7) that
|ρ(t) − ρc |
|Ω − 1| = = [ȧ(t)]−2 , (1.5.1)
ρc
where ρ is the energy density in the universe, and ρc is the critical density for a flat
universe with the same value of the Hubble parameter H(t).
As already mentioned in Section 1.4, the present-day value of Ω is known only roughly,
0.1 <
∼Ω< ∼ 2, or in other words our universe could presently show a fairly sizable departure
from flatness. On the other hand, (ȧ)−2 ∼ t in the early  stages of evolution of a hot
ρ 
universe (see (1.3.14)), so the quantity |Ω − 1| =  − 1 was extremely small. One can
 ρc 
show that in order for Ω to lie in the range 0.1 < <
∼ ∼ 2 now, the early universe must
Ω
2
−59 MP
have had |Ω − 1| < ∼ 10 , so that at T ∼ MP ,
T2
 
ρ 
< −59
|Ω − 1| = − 1 ∼ 10 . (1.5.2)
 
 ρc


This means that if the density of the universe were initially (at the Planck time tP ∼ M−1
P )
−55
greater than ρc , say by 10 ρc , it would be closed, and the limiting value tc would be so
small that the universe would have collapsed long ago. If on the other hand the density
at the Planck time were 10−55 ρc less than ρc , the present energy density in the universe
 18

would be vanishingly low, and the life could not exist. The question of why the energy
density ρ in the early universe was so fantastically close to the critical density (Eq. (1.5.2))
is usually known as the flatness problem.

c. THE TOTAL ENTROPY AND TOTAL MASS PROBLEM. The question here is
why the total entropy S and total mass M of matter in the observable part of the universe,
with Rp ∼ 1028 cm, is so large. The total entropy S is of order (Rp Tγ )3 ∼ 1087 , where
Tγ ∼ 2.7 K is the temperature of the primordial background radiation. The total mass is
given by M ∼ R3p ρc ∼ 1055 g ∼ 1049 tons.
If the universe were open and its density at the Planck time had been subcritical,
say, by 10−55 ρc , it would then be easy to show that the total mass and entropy of the
observable part of the universe would presently be many orders of magnitude lower.
The corresponding problem becomes particularly difficult for a closed universe. We
see from (1.3.15) and (1.3.16) that the total lifetime tc of a closed universe is of order
M−1
P ∼ 10
−43
sec, and this will be a long timespan (∼ 1010 yr) only when the total mass
and energy of the entire universe are extremely large. But why is the total entropy of the
universe so large, and why should the mass of the universe be tens of orders of magnitude
greater than the Planck mass MP , the only parameter with the dimension of mass in the
general theory of relativity? This question can be formulated in a paradoxically simple
and apparently naı̈ve way: Why are there so many different things in the universe?

d. THE PROBLEM OF THE SIZE OF THE UNIVERSE. Another problem associated

with the flatness of the universe is that according to the hot universe theory, the total
size l of the part of the universe currently accessible to observation is proportional to
a(t); that is, it is inversely proportional to the temperature T (since the quantity a T is
practically constant in an adiabatically expanding hot universe — see Section 1.3). This
means that at T ∼ MP ∼ 1019 GeV ∼ 1032 K, the region from which the observable part
of the universe (with a size of 1028 cm) formed was of the order of 10−4 cm in size, or 29
orders of magnitude greater than the Planck length lP ∼ M−1 P ∼ 10
−33
cm. Why, when
the universe was at the Planck density, was it 29 orders of magnitude bigger than the
Planck length? Where do such large numbers come from?
We discuss the flatness problem here in such detail not only because an understanding
of the various aspects of this problem turns out to be important for an understanding of
the difficulties inherent in the standard hot universe theory, but also in order to be able
to understand later which versions of the inflationary universe scenario to be discussed in
this book can resolve this problem.

1.5.3. The problem of the large-scale homogeneity and isotropy of the universe

In Section 1.3, we assumed that the universe was initially absolutely homogeneous and
isotropic. In actuality, or course, it is not completely homogeneous and isotropic even
now, at least on a relatively small scale, and this means that there is no reason to believe
that it was homogeneous ab initio. The most natural assumption would be that the initial
 19

conditions at points sufficiently far from one another were chaotic and uncorrelated [72].
As was shown by Collins and Hawking [73] under certain assumptions, however, the class
of initial conditions for which the universe tends asymptotically (at large t) to a Friedmann
universe (1.3.1) is one of measure zero among all possible initial conditions. This is the
crux of the problem of the homogeneity and isotropy of the universe. The subtleties of
this problem are discussed in more detail in the book by Zeldovich and Novikov [34].

1.5.4. The horizon problem

The severity of the isotropy problem is somewhat ameliorated by the fact that effects
connected with the presence of matter and elementary particle production in an expanding
universe can make the universe locally isotropic [34, 74]. Clearly, though, such effects
cannot lead to global isotropy, if only because causally disjoint regions separated by a
distance greater than the particle horizon (which in the simplest cases is given by Rp ∼ t,
where t is the age of the universe) cannot influence each other. In the meantime, studies
of the microwave background have shown that at t ∼ 105 yr, the universe was quite
accurately homogeneous and isotropic on scales orders of magnitude greater than t, with
temperatures T in different regions differing by less than O(10−4 )T. Inasmuch as the
observable part of the universe presently consists of about 106 regions that were causally
unconnected at t ∼ 105 yr, the probability of the temperature T in these regions being
fortuitously correlated to the indicated accuracy is at most 10−24 –10−30 . It is exceedingly
difficult to come up with a convincing explanation of this fact within the scope of the
standard scenario. The corresponding problem is known as the horizon problem or the
causality problem [48, 56].
There is one more aspect of the horizon problem which will be important for our
purposes. As we mentioned in the earlier discussion of the flatness problem, at the Planck
time tP ∼ M−1 P ∼ 10
−43
sec, when the size (the radius of the particle horizon) of each
causally connected region of the universe was lP ∼ 10−33 cm, the size of the overall region
from which the observable part of the universe formed was of order 10−4 cm. The latter
thus consisted of (1029 )3 ∼ 1087 causally unconnected regions. Why then should the
expansion of the universe (or its emergence from the space-time foam with the Planck
density ρ ∼ M4P ) have begun simultaneously (or nearly so) in such a huge number of
causally unconnected regions? The probability of this occurring at random is close to
exp(−1090 ).

1.5.5. The galaxy formation problem

The universe is of course not perfectly homogeneous. It contains such important inhomo-
geneities as stars, galaxies, clusters of galaxies, etc. In explaining the origin of galaxies, it
has been necessary to assume the existence of initial inhomogeneities [75] whose spectrum
is usually taken to be almost scale-invariant [76]. For a long time, the origin of such
density inhomogeneities remained completely obscure.
 20

1.5.6. The baryon asymmetry problem

The essence of this problem is to understand why the universe is made almost entirely of
matter, with almost no antimatter, and why on the other hand baryons are many orders
nB
of magnitude scarcer than photons, with ∼ 10−9 .
nγ
Over the course of time, these problems have taken on an almost metaphysical flavor.
The first is self-referential, since it can be restated by asking “What was there before
there was anything at all?” or “What was at the time at which there was no space-time
at all?” The others could always be avoided by saying that by sheer good luck, the initial
conditions in the universe were such as to give it precisely the form it finally has now,
and that it is meaningless to discuss initial conditions. Another possible answer is based
on the so-called Anthropic Principle, and seems almost purely metaphysical: we live in
a homogeneous, isotropic universe containing an excess of matter over antimatter simply
because in an inhomogeneous, anisotropic universe with equal amounts of matter and
antimatter, life would be impossible and these questions could not even be asked [77].
Despite its cleverness, this answer is not entirely satisfying, since it explains neither the
nB
small ratio ∼ 10−9 , nor the high degree of homogeneity and isotropy in the universe,
nγ
nor the observed spectrum of galaxies. The Anthropic Principle is also incapable of
explaining why all properties of the universe are approximately uniform over its entire
observable part (l ∼ 1028 cm) — it would be perfectly possible for life to arise if favorable
conditions existed, for example, in a region the size of the solar system, l ∼ 1014 cm.
Furthermore, Anthropic Principle rests on an implicit assumption that either universes
are constantly created, one after another, or there exist many different universes, and that
life arises in those universes which are most hospitable. It is not clear, however, in what
sense one can speak of different universes if ours is in fact unique. We shall return to this
question later and provide a basis for a version of the Anthropic Principle in the context
of inflationary cosmology [57, 78, 79].
The first breach in the cold-blooded attitude of most physicists toward the foregoing
“metaphysical” problems appeared after Sakharov discovered [36] that the baryon asym-
metry problem could be solved in theories in which baryon number is not conserved by
taking account of nonequilibrium processes with C and CP-violation in the very early
universe. Such processes can occur in all grand unified theories [36–38]. The discovery of
a way to generate the observed baryon asymmetry of the universe was considered to be
one of the greatest successes of the hot universe cosmology. Unfortunately, this success
was followed by a whole series of disappointments.

1.5.7. The domain wall problem

√
As we have seen, symmetry is restored in the theory (1.1.5) when T > 2 µ/ λ. As the
temperature drops in an expanding universe, the symmetry is broken. But this symmetry
breaking occurs independently in all causally unconnected regions of the universe, and
therefore in each of the enormous number of such regions comprising the universe at
 21
√
the time of the√symmetry-breaking phase transition, both the field √ ϕ = +µ/ λ and the
field ϕ = −µ/ λ can arise.√Domains filled by the field ϕ = +µ/ λ are separated from
those with the field ϕ − µ/ λ by domain walls. The energy density of these walls turns
out to be so high that the existence of just one in the observable part of the universe
would lead to unacceptable cosmological consequences [41]. This implies that a theory
with spontaneous breaking of a discrete symmetry is inconsistent with the cosmological
data. Initially, the principal theories fitting this description were those with spontaneously
broken CP invariance [80]. It was subsequently found that domain walls also occur in the
simplest version of the SU(5) theory, which has the discrete invariance Φ → −Φ [42], and
in most axion theories [43]. Many of these theories are very appealing, and it would be
nice if we could find a way to save at least some of them.

1.5.8. The primordial monopole problem

Other structures besides domain walls can be produced following symmetry-breaking

phase transitions. For example, in the Higgs model with broken U(1) symmetry and
certain others, strings of the Abrikosov superconducting vortex tube type can occur [81].
But the most important effect is the creation of superheavy t’Hooft–Polyakov magnetic
monopoles [82, 83], which should be copiously produced in practically all of the grand
unified theories [84] when phase transitions take place at Tc1 ∼ 1014 –1015 GeV. It was
shown by Zeldovich and Khlopov [40] that monopole annihilation proceeds very slowly,
and that the monopole density at present should be comparable to the baryon density.
This would of course have catastrophic consequences, as the mass of each monopole is
perhaps 1016 times that of the proton, giving an energy density in the universe about 15
orders of magnitude higher than the critical density ρc ∼ 1029 g/cm3 . At that density,
the universe would have collapsed long ago. The primordial monopole problem is one of
the sharpest encountered thus far by elementary particle theory and cosmology, since it
relates to practically all unified theories of weak, strong, and electromagnetic interactions.

1.5.9. The primordial gravitino problem

One of the most interesting directions taken by modern elementary particle physics is
the study of supersymmetry, the symmetry between fermions and bosons [85]. Here we
will not list all the advantages of supersymmetric theories, referring the reader instead
to the literature [13, 14]. We merely point out that phenomenological supersymmetric
theories, and N = 1 supergravity in particular, may provide a way to solve the mass
hierarchy problem of unified field theories [15]; that is, they may explain why there exist
such drastically differing mass scales MP ≫ MX ∼ 1015 GeV and MX ≫ mW ∼ 102 GeV.
One of the most interesting attempts to resolve the mass hierarchy problem for N = 1
supergravity is based on the suggestion that the gravitino (the spin-3/2 superpartner of
the graviton) has mass m3/2 ∼ mW ∼ 102 GeV [15]. It has been shown [86], however, that
gravitinos with this mass should be copiously produced as a result of high-energy particle
collisions in the early universe, and that gravitinos decay rather slowly.
 22

Most of these gravitinos would only have decayed by the later stages of evolution of
the universe, after helium and other light elements had been synthesized, which would
have led to many consequences that are inconsistent with the observations [44, 45]. The
question is then whether we can somehow rescue the universe from the consequences of
gravitino decay; if not, must we abandon the attempt to solve the hierarchy problem?
Some particular models [87] with superlight or superheavy gravitinos manage to avoid
these difficulties. Nevertheless, it would be quite valuable if we could somehow avoid
the stringent constraints imposed on the parameters of N = 1 supergravity by the hot
universe theory.

1.5.10. The problem of Polonyi fields

The gravitino problem is not the only one that arises in phenomenological theories based
on N = 1 supergravity (and superstring theory). The so-called scalar Polonyi fields χ
are one of the major ingredients of these theories [46, 15]. They are relatively low-mass
fields that interact weakly with other fields. At the earliest stages of the evolution of
the universe they would have been far from the minimum of their corresponding effective
potential V(χ). Later on, they would start to oscillate about the minimum of V(χ), and
as the universe expanded, the Polonyi field energy density ρχ would decrease in the same
manner as the energy density of nonrelativistic matter (ρχ ∼ a−3 ), or in other words
much more slowly than the energy density of hot plasma. Estimates indicate that for the
most likely situations, the energy density presently stored in these fields should exceed the
critical density by about 15 orders of magnitude [47, 48]. Somewhat more refined models
give theoretical predictions of the density ρχ that no longer conflict with the observational
data by a factor of 1015 , but only by a factor of 106 [48], which of course is also highly
undesirable.

1.5.11. The vacuum energy problem

As we have already mentioned, the advent of a constant homogeneous scalar field ϕ over
all space simply represents a restructuring of the vacuum, and in some sense, space filled
with a constant scalar field ϕ remains “empty” — the constant scalar field does not carry a
preferred reference frame with it, it does not disturb the motion of objects passing through
the space that it fills, and so forth. But when the scalar field appears, there is a change
in the vacuum energy density, which is described by the quantity V(ϕ). If there were no
gravitational effects, this change in the energy density of the vacuum would go completely
unnoticed. In general relativity, however, it affects the properties of space-time. V(ϕ)
enters into the Einstein equation in the following way:
1
Rµν − gµν R = 8 π G Tµν = 8 π G (T̃µν + gµν V(ϕ)) , (1.5.3)
2
where Tµν is the total energy-momentum tensor, T̃µν is the energy-momentum tensor of
substantive matter (elementary particles), and gµν V(ϕ) is the energy-momentum tensor
 23

of the vacuum (the constant scalar field ϕ). By comparing the usual energy-momentum
tensor of matter
ρ
 
 −p 
T̃µ ν =  (1.5.4)
 
−p

 
−p
with gµ ν V(ϕ), one can see that the “pressure” exerted by the vacuum and its energy
density have opposite signs, p = −ρ = −V(ϕ).
The cosmological data imply that the present-day vacuum energy density ρvac is not
much greater in absolute value than the critical density ρc ∼ 10−29 g/cm3 :
−29
|ρvac | = |V(ϕ0 )| <
∼ 10 g/cm3 . (1.5.5)

This value of V(ϕ) was attained as a result of a series of symmetry-breaking phase

transitions. In the SU(5) theory, after the first phase transition SU(5) → SU(3) ×SU(2) ×
U(1), the vacuum energy (the value of V(ϕ)) decreased by approximately 1080 g/cm3 .
After the SU(3)×SU(2)×U(1) → SU(3)×U(1) transition, it was reduced by about another
1025 g/cm3 . Finally, after the phase transition that formed the baryons from quarks, the
vacuum energy again decreased, this time by approximately 1014 g/cm3 , and surprisingly
enough after all of these enormous drops, it turned out to equal zero to an accuracy of
±10−29 g/cm3 ! It seems unlikely that the complete (or almost complete) cancellation of
the vacuum energy should occur merely by chance, without some deep physical reason.
The vacuum energy problem in theories with spontaneous symmetry breaking [88] is
presently deemed to be one of the most important problems facing elementary particle
theories.
The vacuum energy density multiplied by 8 π G is usually called the cosmological
constant Λ [89]; in the present case, Λ = 8 π G V(ϕ) [88]. The vacuum energy problem is
therefore also often called the cosmological constant problem.
Note that by no means do all theories ensure, even in principle, that the vacuum
energy at the present epoch will be small. This is one of the most difficult problems
encountered in Kaluza–Klein theories based on N = 1 supergravity in 11-dimensional
space [16]. According to these theories, the vacuum energy would now be of order −M4P ∼
−1094 g/cm−3 . On the other hand, indications that the vacuum energy problem may be
solvable in superstring theories [17] have stimulated a great deal of interest in the latter.

1.5.12. The problem of the uniqueness of the universe

The essence of this problem was most clearly enunciated by Einstein, who said that “we
wish to know not just the structure of Nature (and how natural phenomena are played
out), but insofar as we can, we wish to attain a daring and perhaps utopian goal — to
learn why Nature is just the way it is, and not otherwise” [90]. As recently as a few
years ago, it would have seemed rather meaningless to ask why our space-time is four-
dimensional, why there are weak, strong, and electromagnetic interactions and no others,
 24

e2
why the fine-structure constant α = equals 1/137, and so on. Of late, however, our
4π
attitude toward such questions has changed, since unified theories of elementary particles
frequently provide us with many different solutions of the relevant equations that in
principle could describe our universe.
In theories with spontaneous symmetry breaking, for example, the effective potential
will often have
√ several local minima — in the theory (1.1.5), for instance, there are two,
at ϕ = ±µ/ λ. In the minimal supersymmetric SU(5) grand unification theory, there
are three local minima of the effective potential for the field Φ that have nearly the same
depth [91]. The degree of degeneracy of the effective potential in supersymmetric theories
(the number of different types of vacuum states having the same energy) becomes even
greater when one takes into account other Higgs fields H which enter into the theory [92].
The question then arises as to how and why we come to be in a minimum in which
the broken symmetry is SU(3) × U(1) (this question becomes particularly complicated if
we recall that the early high-temperature universe was at an SU(5)-symmetric minimum
Φ = H = 0 [93], and there is no apparent reason for the entire universe to jump to the
SU(3) × U(1) minimum upon cooling).
It is assumed in the Kaluza–Klein and superstring theories that we live in a space
with d > 4 dimensions, but that d − 4 of these dimensions have been compactified — the
radius of curvature of space in the corresponding directions is of order M−1P . That is why
we cannot move in those directions, and space is apparently four-dimensional.
Presently, the most popular theories of that kind have d = 10 [17], but others with
d = 26 [94] and d = 506 [95, 96] have also been considered. One of the most fundamental
questions that comes up in this regard is why precisely d−4 dimensions were compactified,
and not d − 5 or d − 3. Furthermore, there are usually a great many ways to compactify
d − 4 dimensions, and each results in its own peculiar laws of elementary particle physics
in four-dimensional space. A frequently asked question is then why Nature chose just
that particular vacuum state which leads to the strong, weak, and electromagnetic inter-
actions with the coupling constants that we measure experimentally. As the dimension
d of the parent space rises, this problem becomes more and more acute. Thus, it has
variously been estimated that in d = 10 superstring theory, there are perhaps 101500 ways
of compactifying the ten-dimensional space into four dimensions (some of which may lead
to unstable compactification), and there are many more ways to do this in space with
d > 10. The question of why the world that surrounds us is structured just so, and
not otherwise, has therefore lately turned into one of the most fundamental problems of
modern physics.
We could continue this list of problems facing cosmologists and elementary particle
theorists, of course, but here we are only interested in those that bear some relation to
our basic theme.
The vacuum energy problem has yet to be solved definitively. There are many in-
teresting attempts to do so, some of which are based on quantum cosmology and on the
inflationary universe scenario. A solution to the baryon asymmetry problem was proposed
by Sakharov long before the advent of the inflationary universe scenario [36], but the lat-
 25

ter also introduces much that is new [97–99]. As for the other ten problems, they can all
be solved either partially or completely within the framework of inflationary cosmology,
and we now turn to a description of that theory.

1.6 A sketch of the development of the inflationary universe scenario

The main idea underlying all existing versions of the inflationary universe scenario is that
in the very earliest stages of its evolution, the universe could be in an unstable vacuum-
like state having high energy density. As we have already noted in the preceding section,
the vacuum pressure and energy density are related by Eq. (1.5.4), p = −ρ. This means,
according to (1.3.8), that the vacuum energy density does not change as the universe
expands (a “void” remains a “void”, even if it has weight). But (1.3.7) then implies
that at large times t, the universe in an unstable vacuum state ρ > 0 should expand
exponentially, with
a(t) = H−1 cosh H t (1.6.1)
for k = +1 (a closed Friedmann universe),

a(t) = H−1 eHt (1.6.2)

for k = 0 (a flat universe), and

a(t) = H−1 sinh H t (1.6.3)

s s
8π 8πρ
for k = −1 (an open universe). Here H = Gρ = . More generally, during
3 3 M2P
expansion the magnitude of H in the inflationary universe scenario changes, but very
slowly,
Ḣ ≪ H2 . (1.6.4)
Over a characteristic time ∆t = H−1 there is little change in the magnitude of H, so that
one may speak of a quasiexponential expansion of the universe,
Z t 
a(t) = a0 exp H(t) dt ∼ a0 eHt , (1.6.5)
0

or of a quasi-de Sitter stage in its expansion; just this regime of quasiexponential expansion
is known as inflation.
Inflation comes to an end when H begins to decrease rapidly. The energy stored in
the vacuum-like state is then transformed into thermal energy, and the universe becomes
extremely hot. From that point onward, its evolution is described by the standard hot
universe theory, with the important refinement that the initial conditions for the expansion
stage of the hot universe are determined by processes which occurred at the inflationary
stage, and are practically unaffected by the structure of the universe prior to inflation. As
 26

we shall demonstrate below, just this refinement enables us to solve many of the problems
of the hot universe theory discussed in the preceding section.
The space (1.6.1)–(1.6.3) was first described in the 1917 papers of de Sitter [100],
well before the appearance of Friedmann’s theory of the expanding universe. However,
de Sitter’s solution was obtained in a form differing from (1.6.1)–(1.6.3), and for a long
time its physical meaning was somewhat obscure. Before the advent of the inflationary
universe scenario, de Sitter space was employed principally as a convenient staging area
for developing the methods of general relativity and quantum field theory in curved space.
The possibility that the universe might expand exponentially during the early stages
of its evolution, and be filled with superdense matter with the equation of state p = −ρ,
was first suggested by Gliner [51]; see also [101–103]. When they appeared, however,
these papers did not arouse much interest, as they dealt mainly with superdense baryonic
ρ
matter, which, as we now believe, has an equation of state close to p = , according to
3
asymptotically free theories of weak, strong, and electromagnetic interactions.
It was subsequently realized that the constant (or almost constant) scalar field ϕ
appearing in unified theories of elementary particles could play the role of a vacuum state
with energy density V(ϕ) [88]. The magnitude of the field ϕ in an expanding universe
depends on the temperature, and at times of phase transitions that change ϕ, the energy
stored in the field is transformed into thermal energy [21–24]. If, as sometimes happens,
the phase transition takes place from a highly supercooled metastable vacuum state, the
total entropy of the universe can increase considerably afterwards [23, 24, 104], and in
particular, a cold Friedmann universe can become hot. The corresponding model of the
universe was developed by Chibisov and the present author (in this regard, see [24, 105]).
In 1979–80, a very interesting model of the evolution of the universe was proposed by
Starobinsky [52]. His model was based on the observation of Dowker and Critchley [106]
that the de Sitter metric is a solution of the Einstein equations with quantum corrections.
Starobinsky noted that this solution is unstable, and after the initial vacuum-like state
decays (its energy density is related to the curvature of space R), de Sitter space transforms
into a hot Friedmann universe [52].
Starobinsky’s model proved to be an important step on the road towards the infla-
tionary universe scenario. However, the principal advantages of the inflationary stage had
not yet been recognized at that time. The main objective pursued in [52] was to solve the
problem of the initial cosmological singularity. The goal was not reached at that time,
and the question of initial conditions for the model remained unclear. In that model, fur-
thermore, the density inhomogeneities that appeared after decay of de Sitter space turned
out to be too large [107]. All of these considerations required that the foundations of the
model be significantly altered [108–110]. In its modified form, the Starobinsky model has
become one of the most actively developed versions of the inflationary universe scenario
(or, to be more precise, the chaotic inflation scenario; see below).
The necessity of considering models of the universe with a stage of quasiexponential
expansion was fully recognized only after the work of Guth [53], who suggested using the
exponential expansion (inflation) of the universe in a supercooled vacuum state ϕ = 0
 27

to solve three of the problems discussed in Section 1.5, namely the flatness problem, the
horizon problem, and the primordial monopole problem (a similar possibility for solving
the flatness problem was independently suggested by Lapchinsky, Rubakov, and Veryaskin
[111]). The scenario suggested by Guth was based on three fundamental propositions:
1. The universe initially expands in a state with superhigh temperature and restored
symmetry, ϕ(T) = 0.
2. One considers theories in which the potential V(ϕ) retains a local minimum at
ϕ = 0 even at a low temperature T. As a result, the evolving universe remains in the
supercooled metastable state ϕ = 0 for a long time. Its temperature in this state falls
off, the energy-momentum tensor gradually becomes equal to Tµν = gµν V(0), and the
universe expands exponentially (inflates) for a long time.
3. Inflation continues until the end of a phase transition to a stable state ϕ0 6= 0. This
phase transition proceeds by forming bubbles containing the field ϕ = ϕ0 . The universe
heats up due to bubble-wall collisions, and its subsequent evolution is described by the
hot universe theory.
The exponential expansion of the universe in stage (2) is introduced to make the term
k 8πG
2
in the Einstein equation (1.3.7) vanishingly small as compared with ρ, i.e., in
a 3
order to make the universe flatter and flatter. This same process is invoked to ensure
that the observable part of the universe, some 1028 cm in size, came about as the result
of inflation of a very small region of space that was initially causally connected. In this
scenario, monopoles are created at places where the walls of several exponentially large
bubbles collide, and they therefore have exponentially low density.
The main idea behind the Guth scenario is very simple and extremely attractive.
As noted by Guth himself [53], however, collisions of the walls of very large bubbles
should lead to an unacceptable destruction of homogeneity and isotropy in the universe
after inflation. Attempts to improve this situation were unsuccessful [112, 113] until
cosmologists managed to surmount a certain psychological barrier and renounce all three
of the aforementioned assumptions of the Guth scenario, while retaining the idea that the
universe might have undergone inflation during the early stages of its evolution.
The invention of the so-called new inflationary universe scenario [54, 55] marked the
departure from assumptions (2) and (3). This scenario is based on the fact that inflation
can occur not only in a supercooled state ϕ = 0, but also during the process of growth
of the field ϕ if this field increases to its equilibrium value ϕ0 slowly enough, so that the
time t for ϕ to reach the minimum of V(ϕ) is much longer than H−1 . This condition can
be realized if the effective potential of the field ϕ has a sufficiently flat part near ϕ = 0. If
inflation during the stage when ϕ is rolling downhill is large enough, the walls of bubbles
of the field ϕ (if they are formed) will, after inflation, be separated from one another by
much more than 1028 cm, and will not engender any inhomogeneities in the observable
part of the universe. In this scenario, the universe is heated after inflation not because
of collisions between bubble walls, but because of the creation of elementary particles by
the classical field ϕ, which executes damped oscillations about the minimum of V(ϕ).
The new inflationary universe scenario is free of the major shortcomings of the old
 28

scenario. In the context of this scenario, it is possible to propose solutions not just to
the flatness, horizon, and primordial monopole problems, but also to the homogeneity
and isotropy problems, as well as many of the others referred to in Section 1.5. It has
been found, in particular, that at the time of inflation in this scenario, density inhomo-
geneities are produced with a spectrum that is virtually independent of the logarithm
of the wavelength (a so-called flat, scale-free, or Harrison–Zeldovich spectrum [214, 76]).
This marked an important step on the road to solving the problem of the origin of the
large-scale structure of the universe.
The successes of the new inflationary universe scenario were so impressive that even
now, many scientists who speak of the inflationary universe scenario mean this new sce-
nario [54, 55]. In our opinion, however, this scenario is still far from perfect; there are at
least three problems that stand in the way of its successful implementation:
1. The new scenario requires a realistic theory of elementary particles in which the
effective potential satisfies many constraints that are rather unnatural. For example, the
potential V(ϕ) must be very close to flat (V(ϕ) ≈ const) for values of the field close to
λ
ϕ = 0. If for instance, the behavior of V(ϕ) at small ϕ is close to V(0) − ϕ4 , then in
4
order for density inhomogeneities generated at the time of inflation to have the required
amplitude
δρ
∼ 10−4 –10−5 , (1.6.6)
ρ
the constant λ must be extremely small [114],

λ ∼ 10−12 –10−14 . (1.6.7)

On the other hand, the curvature of the effective potential V(ϕ) near its minimum at
ϕ = ϕ0 must be great enough to make the field ϕ oscillate at high frequency after inflation,
thereby heating the universe to a rather high temperature T. It has turned out to be rather
difficult to suggest a natural yet realistic theory of elementary particles that satisfies all
the necessary requirements.
2. The second problem is related to the fact that the weakly interacting field ϕ (see
(1.6.7)) is most likely not to be in a state of thermodynamic equilibrium with the other
fields present in the early universe. But even if it were in equilibrium, if λ is small, high-
temperature corrections to V(ϕ) of order λ T2 ϕ2 cannot alter the initial value of the field
ϕ and make it zero in the time between the birth of the universe and the assumed start
of inflation [115, 116].
3. Yet another problem relates to the fact that in both the old and new scenarios,
inflation will only begin when the temperature of the universe has dropped sufficiently
far, T4 <∼ V(0). However, the condition (1.6.6) implies not only the constraint (1.6.7)
on λ, but also (in most models) a constraint on the value of V(ϕ) in the last stages of
inflation, which in the new inflationary universe scenario is practically equal to V(0) [116,
117]:
−13 4
V(0) <∼ 10 MP . (1.6.8)
 29

This means that inflation will start when T2 < −7 2

∼ 10 MP , i.e., at a time t −1following the
beginning of expansion of the universe that exceeds the Planck time tP ∼ MP (1.3.20) by
6 orders of magnitude. But for a hot, closed universe to live that long, its total entropy
must at the very outset be greater than S ∼ 109 (1.3.16). Thus, the flatness problem for
a closed universe has not been solved [116], either in the context of the Guth scenario or
the new inflationary universe scenario. One could look upon this result as an argument
in favor of the universe being either open or flat. We think, however, that this is not a
problem of the theory of a closed universe; rather, it is just one more shortcoming of the
new inflationary universe scenario.
Fortunately, there is another version of the inflationary universe scenario, the so-called
chaotic inflation scenario [56, 57], which does not share these problems. Rather than
being based on the theory of high-temperature phase transitions, it is simply concerned
with the evolution of a universe filled with a chaotically (or almost chaotically — see
below) distributed scalar field ϕ. In what follows, we will discuss this scenario and the
considerable changes that have taken place in recent years in our ideas about the early
stages of the evolution of the universe, and about its large-scale structure.

1.7 The chaotic inflation scenario

We will now illustrate the basic idea of the chaotic inflation scenario with an example
drawn from the simplest theory of the scalar field ϕ minimally coupled to gravity, with
the Lagrangian
1
L = ∂µ ϕ ∂ µ ϕ − V(ϕ) . (1.7.1)
2
We shall also assume that when ϕ > ∼ MP , the potential V(ϕ) rises more slowly than
6ϕ
 
(approximately) exp . In particular, this requirement is satisfied by any potential
MP
that follows a power law for ϕ >
∼ MP :
λ ϕn
V(ϕ) = , (1.7.2)
n MPn−4

n > 0, 0 < λ ≪ 1.
In order to study the evolution of a universe filled with a scalar field ϕ, we must
somehow set the initial values of the field and its derivatives at different points in space,
and also specify the topology of the space and its metric in a manner consistent with the
initial conditions for ϕ. We might assume, for example, that from the very beginning, the
field ϕ over all space is in the equilibrium state ϕ = ϕ0 corresponding to a minimum of
V(ϕ). But this would be even more unconvincing than assuming that the whole universe
is perfectly uniform and isotropic from the very beginning. Actually, regardless of whether
the universe was originally hot or its dynamical behavior was determined solely by the
classical field ϕ, at a time t ∼ tP ∼ M−1
P after the singularity (or after the quantum birth
 30

of the universe — see below) the energy density ρ (and consequently the value of V(ϕ))
was determined only to accuracy O(M4P ) by virtue of the Heisenberg uncertainty principle.
Assuming that the field ϕ initially taken value ϕ = ϕ0 is therefore no more plausible than
assuming it taken any other value with

∂0 ϕ ∂ 0 ϕ <
∼ M4P , (1.7.3)
∂i ϕ ∂ i ϕ <
∼ M4P , i = 1, 2, 3 , (1.7.4)
V(ϕ) <
∼ M4P , (1.7.5)
R2 <
∼ M4P . (1.7.6)

The last of these inequalities is taken to mean that invariants constructed from the curva-
ture tensor Rµναβ are less than corresponding powers of the Planck mass (Rµναβ Rµναβ < 4
∼ MP ,
Rµ ν Rν α Rα µ < 6
∼ MP , etc.). It is usually assumed that the first instant at which the forego-
ing conditions hold is the instant after which the region of the universe under consideration
can be described as a classical space-time (in nonstandard versions of gravitation theory,
the corresponding conditions may generally differ from (1.7.3)–(1.7.6)). It is precisely this
instant after which one can speak of specifying the initial distribution of a classical scalar
field ϕ in a region of classical space-time.
Since there is absolutely no a priori reason to expect that ∂µ ϕ ∂ µ ϕ ≪ M4P , R2 ≪ M4P ,
or V(ϕ) ≪ M4P , it seems reasonable to suppose that the most natural initial conditions at
the moment when the classical description of the universe first becomes feasible are

∂0 ϕ ∂ 0 ϕ ∼ M4P , (1.7.7)
∂i ϕ ∂ i ϕ ∼ M4P , i = 1, 2, 3 , (1.7.8)
V(ϕ) ∼ M4P , (1.7.9)
R2 ∼ M4P . (1.7.10)

We shall return to a discussion of initial conditions in the early universe more than once
in the main body of this book, but for the moment, we will attempt to understand the
consequences of the assumption made above [56, 118].
Investigation of the expansion of the universe with initial conditions (1.7.7)–(1.7.10)
is still an extremely complicated problem, but there is a simplifying circumstance that
carries one a long way toward a solution. Specifically, we are most interested in studying
the possibility that regions of the universe will form that look like part of an exponentially
expanding Friedmann universe. As we have already noted in Section 1.4, the latter is a
de Sitter space, with only a small part of that space, of radius H−1 , being accessible to
a stationary observer. This observer sees himself as surrounded by a black hole situated
at a distance H−1 , corresponding to the event horizon of the de Sitter space. It is well
known that nothing entering a black hole can reemerge, nor can anything that has been
captured affect physical processes outside the black hole. This assertion (with certain
qualifications that need not concern us here) is known as the theorem that “a black
hole has no hair” [119]. There is an analogous theorem for de Sitter space as well:
all particles and other inhomogeneities within a sphere of radius H−1 will have left that
 31

sphere (crossed the event horizon) by a time of order H−1 , and will have no effect on events
taking place within the horizon (de Sitter space “has no hair” [120, 121]). As a result,
the local geometrical properties of an expanding universe with energy-momentum tensor
Tµν ≈ gµν V(ϕ) approach those of de Sitter space at an exponentially high rate; that is,
the universe becomes homogeneous and isotropic, and the total size of the homogeneous
and isotropic region rises exponentially [120–122].
In order for such behavior to be feasible, the size of the domain within which the
expansion takes place must exceed 2 H−1 . When V(ϕ) ∼ M4P , the horizon is as close as it
can be, with H−1 ∼ M−1 P ; that is, we are dealing with the smallest domains that can still
be described in terms of classical space-time. Moreover, it is necessary that expansion be
approximately exponential in order for the event horizon H−1 (t) to recede slowly enough,
and for inhomogeneities at the time of expansion to escape beyond the horizon, without
engendering any back influence on the expansion taking place within the horizon. This
condition will be satisfied if Ḣ ≪ H2 , and this is just the situation during the stage of
inflation.
Thus, to assess the possibility of inflationary regions arising in a universe with initial
conditions (1.7.7)–(1.7.10), it is sufficient to consider whether inflationary behavior could
arise at the Planck epoch in an isolated domain of the universe, with the minimum size l
that could still be treated in terms of classical space-time, l ∼ H−1 (ϕ) ∼ M−1 P .
The significance of (1.7.9) is that the typical initial value ϕ0 of the field ϕ in the early
λ
universe is exceedingly large. For example, in a theory with V(ϕ) = ϕ4 and λ ≪ 1,
4
ϕ0 (x) ∼ λ−1/4 MP ≫ MP . (1.7.11)

According to (1.7.4) and (1.7.11), in any region whose size is of the order of the event
horizon H−1 (ϕ) ∼ M−1 P , the field ϕ0 (x) changes by a relatively insignificant amount,
∆ϕ ∼ MP ≪ ϕ0 . In each such domain, as we have said, the evolution of the field proceeds
independently of what is happening in the rest of the universe.
Let us consider such a region of the universe having initial size O(M−1 P ), in which
µ
∂µ ϕ ∂ ϕ and the squares of the components of the curvature tensor Rµναβ , which are re-
sponsible for the inhomogeneity and anisotropy of the universe,7 are several times smaller
than V(ϕ) ∼ M4P . Since all these quantities are typically of the same order of magnitude
according to (1.7.7)–(1.7.10), the probability that regions of the specified type do exist
should not be much less than unity. The subsequent evolution of such regions turns out
to be extremely interesting.
In fact, the relatively low degree of anisotropy and inhomogeneity of space in such
regions enables one to treat each of them as being a locally Friedmann space, with a
7
Note that the quantities ∂µ ∂ µ ϕ and R2 cannot exceed V(ϕ) in one small part of the region considered
and be less than V(ϕ) in another, since it is not possible to subdivide classical space into parts less than
M−1P in size and consider the classical field ϕ separately in each of these parts, due to the large quantum
fluctuations of the metric at this scale.
 32

metric of the type (1.3.1), governed by Eq. (1.3.7):

 2
ϕ̇2 (∇ϕ)2
!
2 k ȧ k 8π
H + 2 ≡ + 2 = + + V(ϕ) . (1.7.12)
a a a 3 M2P 2 2

At the same time, the field ϕ satisfies the equation

ȧ 1 dV
ϕ = ϕ̈ + 3 ϕ̇ − 2 ∆ϕ = − , (1.7.13)
a a dϕ
where is the covariant d’Alembertian operator, and ∆ is the Laplacian in three-
dimensional space with the time-independent metric

dr 2
dl2 = + r 2 (dθ2 + sin2 θ dϕ2 ) . (1.7.14)
1 − k r2
For a sufficiently ! uniform and slowly varying field ϕ
dV
ϕ̇2 , (∇ϕ)2 ≪ V; ϕ̈ ≪ , Eqs. (1.7.12) and (1.7.13) reduce to
dϕ
 2
2 k ȧ k 8π
H + 2 ≡ + 2 = V(ϕ) , (1.7.15)
a a a 3 M2P
dV
3 H ϕ̇ = − . (1.7.16)
dϕ

It is not hard to show that if the universe is expanding (ȧ > 0) and, as we have said,
the initial value for ϕ satisfies (1.7.11), then the solution of the system of equations
(1.7.15) and (1.7.16) rapidly proceeds to its asymptotic limit of quasiexponential expan-
k
sion (inflation), whereupon the term 2 in (1.7.15) can be neglected. Such behavior is
a
8 π V(ϕ)
understandable, inasmuch as Eq. (1.7.15) tells us that when a2 is large, H2 = .
3 M2P
It then follows from (1.7.16) that
!2
1 2 M2P dV
ϕ̇ = . (1.7.17)
2 48 π V dϕ

Hence, for V(ϕ) ∼ ϕn , we have that

1 2 n2 M2P
ϕ̇ = V(ϕ) , (1.7.18)
2 48 π ϕ2
1 2
i.e., that ϕ̇ ≪ V(ϕ) when
2
n
ϕ≫ √ MP . (1.7.19)
4 3π
 33

MP4

− 1/4
0 MP λ MP ϕ

Figure 1.5: Evolution of a homogeneous classical scalar field ϕ in a theory with V(ϕ) =
λ 4
ϕ , neglecting quantum fluctuations of the field. When ϕ > λ−1/4 MP , the energy
4
density of the field ϕ is greater than the Planck density, and the evolution of the universe
MP
cannot be described classically. When <
∼ ϕ<∼ λ−1/4 MP , the field ϕ slowly decreases,
3
MP
and the universe then expands quasiexponentially (inflates). When ϕ < ∼ 3 , the field ϕ
oscillates rapidly about the minimum of V(ϕ), and transfers its energy to the particles
produced thereby (reheating of the universe).

This means that for large ϕ, the energy-momentum tensor Tµν of the field ϕ is de-
termined almost entirely by the quantity gµν V(ϕ), or in other words, p ≈ −ρ, and the
universe expands quasiexponentiallly. Because of the fact that when ϕ ≫ MP the rates
at which the field ϕ and potential
! V(ϕ) vary are much less than the rate of expansion of
ϕ̇ H
the universe ≪ H, Ḣ ≪ H2 , over time intervals ∆t <
∼ ≫ H−1 the universe looks
ϕ Ḣ
approximately like de Sitter space with the expansion law

a(t) ∼ eHt (1.7.20)

where the quantity v

u 8 π V(ϕ)
u
H(ϕ(t)) = t (1.7.21)
3 M2P
decreases slowly with time [56].
Under these conditions, the behavior of the field ϕ(t) (see Fig. 1.5) is
 s 
λ
ϕ(t) = ϕ0 exp − MP t (1.7.22)
6π
 34

λ 4
for a theory with V(ϕ) = ϕ , and
4
s
n n λ 3− n2
 
2− n 2− n
ϕ(t) 2 = ϕ0 2
+t 2− M (1.7.23)
2 24 π P
m2 ϕ2
for V(ϕ) ∼ ϕn (1.7.2), with n 6= 4. In particular, for a theory with V(ϕ) = (i.e.,
2
for n = 2 and λ M2P = m2 ),
m MP
ϕ(t) = ϕ0 − √ t. (1.7.24)
2 3π
Meanwhile, the behavior of the scale factor of the universe is given by the general equation
4π
a(t) = a0 exp (ϕ2 − ϕ2 (t)) , (1.7.25)
n M2P 0
which yields Eq. (1.7.20) for sufficiently small t. Making use of the estimate (1.7.18), one
n
can easily see that this regime (the inflation regime) ends when ϕ < ∼ 12 MP . If ϕ0 ≫ MP ,
then (1.7.24) implies that the overall inflation factor P for the universe at that time is
!
4π 2
P ≈ exp ϕ . (1.7.26)
n M2P 0
According to (1.7.26), then, the degree of inflation is small for small initial values of
the field ϕ, and it grows exponentially with increasing ϕ0 . This means that most of the
physical volume of the universe comes into being not by virtue of the expansion of regions
which initially, and randomly, contained a small field ϕ (or a markedly inhomogeneous
and rapidly varying field ϕ that failed to lead to exponential expansion of the universe),
but as a result of the inflation of regions of a size exceeding the radius of the event
horizon H−1 (ϕ) which were initially filled with a sufficiently homogeneous, slowly varying,
extremely large field ϕ = ϕ0 . The only fundamental constraint on the magnitude of the
4
homogeneous, slowly varying field ϕ is V(ϕ) < ∼ MP (1.7.5). As we have already mentioned,
−1
the probability that domains of size ∆l > −1
∼ H (ϕ) ∼ MP exist in the early universe with
−1
ϕ̇2 , (∇ϕ)2 < ∼ V(ϕ) ∼ M P should not be significantly suppressed. In conjunction with
(1.7.26), this leads one to believe that most of the physical volume of the present-day
universe came into being precisely as a result of the exponential expansion of regions of
the aforementioned type.
If in the initial state, as we are assuming,
λ ϕn0
V(ϕ0 ) ∼ ∼ M4P , (1.7.27)
n MPn−4
the inflation factor of the corresponding region is
 !− 2 
4π λ n
P ∼ exp   . (1.7.28)
n n
 35

a O
F
C
hot universe

reheating

inflation

O
F
? C
0 tP ∼ 10–43 sec t ∼ 10 –35 sec t0 ∼ 1017 sec t

Figure 1.6: The lighter set of curves depicts the behavior of the size of the hot universe
(or more precisely, its scale factor) for three Friedmann models: open (O), flat (F), and
closed (C). The heavy curves show the evolution of an inflationary region of the universe.
Because of quantum gravitational fluctuations, the classical description of the expansion
of the universe cannot be valid prior to t ∼ tP = M−1 P ∼ 10
−43
sec after the Big Bang at
t = 0 (or after the start of inflation in the given region). In the simplest models, inflation
continues for approximately 10−35 sec. During that time, the inflationary region of the
7 14
universe grows by a factor of from 1010 to 1010 . Reheating takes place afterwards, and
the subsequent evolution of the region is described by the hot universe theory.

λ 4
In particular, for a ϕ theory
4
!
2π
P ∼ exp √ , (1.7.29)
λ
m2 ϕ2
while for an theory,
2
4π M2P
P ∼ exp . (1.7.30)
m2
After the field ϕ decreases in magnitude to a value of order MP (1.7.18), the quantity
H, which plays the role of a coefficient of friction in Eq. (1.7.13), is no longer large enough
to prevent the field ϕ from rapidly rolling down to the minimum of the effective potential.
The field ϕ starts its oscillations near the minimum of V(ϕ), and its energy is transferred
to the particles that are created as a result of these oscillations. The particles thus created
collide with one another, and approach a state of thermodynamic equilibrium — in other
words, the universe heats up [53, 123, 124] (see Fig. 1.6).
If this reheating of the universe occurs rapidly enough (during the time ∆t < −1
∼ H (ϕ ∼
MP )), virtually all of the energy from the oscillating field will be transformed into thermal
 36

energy, and the temperature of the universe after reheating will be given by
π 2 N(TR ) 4 n
 
TR ∼ V ϕ ∼ MP . (1.7.31)
30 12
λ 4
For example, with N(T) ∼ 103 for the V(ϕ) = ϕ theory, TR = c λ1/4 MP , where
4
c = O(10−1 ). In many realistic versions of the inflationary scenario, however, the temper-
atureof the universe
 after reheating is found to be many orders of magnitude lower than
1/4 n
V ϕ∼ MP because of the inefficiency of the reheating process that results from
12
the weak interaction of the field ϕ with itself and with other fields (see below).
One circumstance that is especially important is that both the value and the behavior
of the field ϕ near ϕ ∼ MP are essentially independent of its initial value ϕ0 when ϕ0 ≫
MP ; that is, the initial temperature of the universe after reheating depends neither on the
initial conditions during the inflationary stage nor its duration, etc. The only parameter
that changes during inflation is the scale factor, which grows exponentially in accordance
with (1.7.28)–(1.7.30). This is precisely the circumstance that enables us to solve the
majority of the problems recounted in Section 1.5.
First of all, let us discuss the problems of the flatness, homogeneity, and isotropy of
space. Note that during the quasiexponential expansion of the universe, the right-hand
k
side of Eq. (1.7.12) decreases very slowly, while the term 2 on the left-hand side falls off
a
exponentially. Thus, the local difference between the three-dimensional geometry of the
universe and the geometry of flat space also falls off exponentially, although the global
topological properties of the universe remain unchanged. To solve the flatness problem,
it is necessary that during inflation a region of initial size ∆l ∼ M−1 P ∼ 10
−33
cm grow
30
by a factor of roughly 10 (see Section 1.5). This condition is amply satisfied in most
specific realizations of the chaotic inflation scenario (see below), and in contrast to the
situation in the new inflationary universe scenario, inflation can begin in the present
scenario at energy densities as high as one might wish, and arbitrarily soon after the
universe starts to expand, i.e., prior to the moment when a closed universe starts to
recollapse. After a closed universe has passed through its inflationary stage, its size (and
therefore its lifetime) becomes exponentially large. The flatness problem in the chaotic
inflation scenario is thereby solved, even if the universe is closed.
The solution of the flatness problem in this scenario has a simple, graphic interpreta-
tion: when a sphere inflates, its topology is unaltered, but its geometry becomes flatter
(Fig. 1.7). The analogy is not perfect, but it is reasonably useful and instructive. It
is clear, for instance, that if the Himalayas were drastically stretched horizontally while
their height remained fixed, we would find a plain in place of the mountains. The same
thing happens during inflation of the universe. Thus, for example, rapid inflation inhibits
time-dependent changes in the amplitude of the field ϕ (the term 3 H ϕ̇ in (1.7.13) plays
the role of viscous damping), i.e., the distribution of the field ϕ in coordinates r, θ, ϕ is
“frozen in.” At the same time, the overall scale of the universe a(t) grows exponentially,
so that the distribution of the classical field ϕ per unit physical volume approaches spatial
 37

Figure 1.7: When an object increases enormously in size, its surface geometry becomes
almost Euclidean. This effect is fundamental to the solution of the flatness, homogeneity,
and isotropy problems in the observable part of the universe, by virtue of the exponentially
rapid inflation of the latter.

uniformity at an exponential rate, ∂i ϕ ∂ i ϕ → 0. At the same time, the energy-momentum

tensor rapidly approaches gµν V(ϕ) (to within small corrections ∼ ϕ̇2 ) curvature tensor
acquires the form

Rµναβ = H2 (gµν gαβ − gµβ gνα ) , (1.7.32)

Rµν = 3 H2 gµν , (1.7.33)
32 π
R = 12 H2 = V(ϕ) , (1.7.34)
M2P
and the difference between the properties of this domain of the universe and those of
the homogeneous, isotropic Friedmann universe (1.3.1) becomes exponentially small (in
complete accord with the “no hair” theorem for de Sitter space). After inflation, this
homogeneous and isotropic domain becomes exponentially large. This explains the ho-
mogeneity and isotropy of the observable part of the universe [54–56, 120–122].
The stretching of the scales of all inhomogeneities leads to an exponential decrease in
the density of monopoles, domain walls, gravitinos, and other entities produced before or
during inflation. If TR , the temperature of the universe after reheating, is not high enough
 38

to produce monopoles, domain walls, and gravitinos again, the corresponding problems
disappear.
Simultaneously with the smoothing of the original inhomogeneities and ejection of
monopoles and domain walls beyond the limits of the observable universe, inflation itself
gives rise to specific large-scale inhomogeneities [107, 114, 125]. The theory of this phe-
nomenon is quite complicated; it will be considered in Section 7.5. Physically, the reason
for the appearance of large-scale inhomogeneities in an inflationary universe is related
to the restructuring of the vacuum state resulting from the exponential expansion of the
universe. It is well known that the expansion of the universe often leads to the production
of elementary particles [74]. It turns out that the usual particles are produced at a very
low rate during inflation, but inflation converts short-wavelength quantum fluctuations
δϕ of the field ϕ into long-wavelength fluctuations. In an inflationary universe, short-
wavelength fluctuations of the field ϕ are no different from short-wavelength fluctuations
in the Minkowski space (1.1.13) (a field with momentum k ≫ H does not “feel” the cur-
vature of space). After the wavelength of a fluctuation δϕ exceeds the horizon H−1 in
size, however, its amplitude is “frozen in” (due to the damping term 3 Hϕ̇ in (1.7.13));
that is, the field δϕ stops oscillating, but the wavelength of the field δϕ keeps growing
exponentially. Looked at from the standpoint of conventional scalar field quantization
in a Minkowski space, the appearance of such scalar field configurations may be inter-
preted not as the production of particles of the field ϕ (1.1.13), but as the creation of
an inhomogeneous (quasi)classical field dϕ(x), where the degree to which it can be con-
sidered quasiclassical rises exponentially as the universe expands. One could say that
in a certain sense an inflationary universe works like a laser, continuously generating
waves of the classical field ϕ with wavelength l ∼ k −1 ∼ H−1 . There is an important
difference, however, in that the wavelength of the inhomogeneous classical field δϕ that
is produced then grows exponentially with time. Small-scale inhomogeneities of the field
ϕ that arise are therefore stretched to exponentially large sizes (with their amplitudes
changing very slowly), while new small-scale inhomogeneities δϕ(x) are generated in their
place.
The typical time scale in an inflationary universe is of course ∆t = H−1 . The mean
amplitude of the field δϕ(x) with wavelength l ∼ k −1 ∼ H−1 generated over this period is
[126–128]
H(ϕ)
|δϕ(x)| ∼ . (1.7.35)
2π
Since H(ϕ) varies very slowly during inflation, the amplitude of perturbations of the field
ϕ that are formed over ∆t = H−1 a time will have only weak time dependence. Bearing
in mind, then, that the wavelength l ∼ k −1 of fluctuations δϕ(x) depends exponentially
on the inflation time t, it can be shown that the spectrum of inhomogeneities of the field
ϕ formed during inflation and the spectrum of density inhomogeneities δρ proportional
to δϕ are almost independent of wavelength l (momentum k) on a logarithmic scale.
As we have already mentioned, inhomogeneity spectra of this type were proposed long
ago by cosmologists studying galaxy formation [76, 214]. The theory of galaxy formation
requires, however, that the relative amplitude of density fluctuations with such a spectrum
 39

be fairly low,
δρ(k)
∼ 10−4 –10−5 . (1.7.36)
ρ
δρ λ
At the same time, estimates of the quantity in the V(ϕ) ∼ ϕ4 theory yield [114, 116]
ρ 4
δρ √
∼ 102 λ , (1.7.37)
ρ
whereupon we find that the constant λ should be extremely small,

λ ∼ 10−13 –10−14 , (1.7.38)

exactly as in the new inflationary universe scenario. With this value of λ, the typical
inflation factor for the universe is of order
π 5
P ∼ exp √ ∼ 1010 . (1.7.39)
λ
During inflation, a region of initial size ∆l ∼ lP ∼ M−1
P ∼ 10
−33
cm will grow to
π 105
L ∼ M−1
P exp √ ∼ 10 cm , (1.7.40)
λ
which is many orders of magnitude larger than the observable part of the universe, Rp ∼
1028 cm. According to (1.7.22), the total duration of inflation will be
s
1 6 π −1 1
τ∼ MP ln ∼ 108 M−1
P ∼ 10
−35
sec . (1.7.41)
4 λ λ
The estimates (1.7.39) and (1.7.40) make it clear how the horizon problem is resolved
in the chaotic inflation scenario: expansion began practically simultaneously in different
28
regions of the observable part of the universe with a size l <
∼ 10 cm, since they all came
into being as a result of inflation of a region of the universe no bigger than 10−33 cm,
which started simultaneously to within ∆t ∼ tP ∼ 10−43 sec. The exponential expansion
of the universe makes it causally connected at scales many orders of magnitude greater
than the horizon size in a hot universe, RP ∼ c t.
These results may seem absolutely incredible, especially when one realizes that the
entire observable part of the universe, which according to the hot universe theory has now
been expanding for about 1010 years, is incomparably smaller than a single inflationary
domain which started with the smallest possible initial size, ∆l ∼ lP ∼ M−1 P ∼ 10
−33

cm (1.7.40), and expanded in a matter of 10−37 sec. Here we must again emphasize that
such a rapid increase in the size of the universe is not at variance with the conventional
limitation on the speed at which a signal can propagate, v ≤ c = 1 (see Section 1.4). On
the other hand, it must also be understood that the actual numerical estimates (1.7.39)
δρ
and (1.7.40) depend heavily on the model used. For example, if ∼ 10−5 in the theory
ρ
 40

m2 ϕ2 14 7
with V(ϕ) = , the characteristic inflation factor P becomes 1010 instead of 1010 ;
2
the inflation factor is much smaller in some other models. For our purposes, it will
only be important that after inflation, the typical size of regions of the universe that we
consider become many orders of magnitude larger than the observable part of the universe.
k
Accordingly, the quantity 2 in (1.3.7) will then be many orders of magnitude less than
a
8π
G ρ; i.e., the universe after inflation becomes (locally) indistinguishable from a flat
3
universe. This implies that the density of the universe at the present time must be very
close to the critical value,
ρ
Ω= =1, (1.7.42)
ρc
δρ
to within ∼ 10−3 –10−4 , which is related to local density inhomogeneities in the observ-
ρc
able part of the universe. This is one of the most important predictions of the inflationary
universe scenario, and in principle it can be verified using astronomical observations.
Let us now turn to the problem of reheating of the universe after inflation. For
λ
λ ∼ 10−14 in the ϕ4 theory, the temperature of the universe after reheating, according
4
to (1.7.31), cannot typically exceed

TR ∼ 10−1 λ1/4 MP ∼ 3 · 1014 GeV . (1.7.43)

As a rule, TR actually turns out to be even lower. In the first place,

√ in this theory, the
12
oscillation frequency of the field ϕ near the minimum of V(ϕ) is λ MP ∼ 10 GeV at
most, and in some theories it is impossible to reheat the universe to higher temperatures.
Furthermore, the weakness with which ϕ interacts with other fields retards reheating. As
a result, the oscillation amplitude of the field ϕ decreases as the universe expands, due to
the term 3 H ϕ̇ in the equation of motion for ϕ, and the temperature of the universe after
inflation turns out to be much lower in certain theories than the value in (1.7.43).
Generally speaking, this can lead to some difficulties in treating the baryon asym-
metry problem. Actually, during inflation, any initial baryon asymmetry in the universe
dies out exponentially, and for such asymmetry to arise after inflation becomes not just
aesthetically attractive, as in the usual hot universe theory, but necessary. Moreover, the
mechanism for producing the baryon asymmetry that was proposed in [36–38] and worked
out in the context of grand unified theories is only effective if the temperature T is high
enough that superheavy particles appear in the hot plasma, with their subsequent decay
producing an excess of baryons over antibaryons. Usually, for this to happen, the temper-
ature of the universe must be higher than 1015 GeV, and this is seldom achievable in the
inflationary universe scenario. Fortunately, however, baryon production can also proceed
at much lower temperatures after inflation, due to nonequilibrium processes that take
place during reheating [123]. Moreover, a number of models have recently been suggested
which allow for the onset of baryon asymmetry even if the temperature of the universe af-
ter inflation never exceeds 102 GeV [97–99, 129]. Thus, the inflationary universe scenario
 41

log T

–28
log t

40
} Lepton desert

Decay of baryons

–22 30

Death of the sun –15 20

Emergence of mankind
Birth of the sun
–8 10

–3 0
Formation of baryons
from quarks

2 –10 Symmetry breaking between

weak and electromagnetic
interactions

7 –20

Generation of the baryon

asymmetry of the universe
12 –30
Symmetry breaking between
strong and electroweak
Inflation interactions
–40
Planck time

Figure 1.8: The main stages in the evolution of an inflationary region of the universe.
The time t is measured in seconds from the start of inflation, and the temperature T is
measured in GeV (1 GeV ≈ 1013 K). The typical lifetime of a region of the universe,
between the start of inflation and the region’s collapse (if it exceeds the critical density),
is many orders of magnitude greater than the proton decay lifetime in the simplest grand
unified theories.

can successfully incorporate all basic results of the hot universe theory, and the resulting
theory proves to be free of the main difficulties of the standard hot Big Bang cosmology.
The main stages in the evolution of an inflationary domain of the universe are shown
in Fig. 1.7. The initial and final stages in the development of each individual inflationary
domain depend on the global structure of the inflationary universe, which we will discuss
in Section 1.8.
 42

1.8 The self-regenerating universe

The attentive reader probably already has noticed that in discussing the problems resolved
with the aid of the inflationary universe scenario, we have silently skirted the most impor-
tant one — the problem of the cosmological singularity. We have also said nothing about
the global structure of the inflationary universe, having limited ourselves to statements
to the effect that its local properties are very similar to those of the observable world.
The study of the global structure of the universe and the problem of the cosmological
singularity within the scope of the inflationary universe scenario conceals a number of
surprises. Prior to the advent of this scenario, there was absolutely no reason to suppose
that our universe was markedly inhomogeneous on a large scale. On the contrary, the
astronomical data attested to the fact that on large scales, up to the very size of the entire
δρ
observable part of the universe Rp ∼ 1028 cm, inhomogeneities on the average were at
ρ
most 10−3 . To understand the evolution of the universe, it was in large measure thought to
be sufficient to investigate homogeneous (or slightly inhomogeneous) cosmological models
like the Friedmann model (or anisotropic Bianchi models) [65].
Meanwhile, the results of the preceding section make it clear that the observable part
of the universe is most likely just a minuscule part of the universe as a whole, and it is an
impermissible extrapolation to draw any conclusions about the homogeneity of the latter
based on observations of such a tiny component. On the contrary, an investigation of
the global geometry of the inflationary universe shows that the universe, which is locally
Friedmann, should be completely inhomogeneous on the largest scales, and its global
geometry and dynamical behavior as a whole have nothing in common with the geometry
and dynamics of a Friedmann universe [57, 78, 132, 133].
In order to obtain a simple derivation of this important and somewhat surprising
result, let us consider more carefully the behavior of the scalar field ϕ for the minimal
λ
model (1.7.1) with V(ϕ) = ϕ4 in the chaotic inflation scenario, taking into account
4
long-wave fluctuations of ϕ that arise during inflation [57]. We have from (1.7.21) and
(1.7.22) that in a typical time
s
3 MP
∆t = H−1 (ϕ) = , (1.8.1)
2 π λ ϕ2
the classical homogeneous field ϕ decreases by

M2P
∆ϕ = . (1.8.2)
2πϕ

During this same period of time, according to (1.7.36), inhomogeneities of the field ϕ are
generated having wavelength l > −1
∼ H and mean amplitude
s
H(ϕ) λ ϕ2
|δϕ(x)| ≈ = . (1.8.3)
2π 6 π MP
 43

MP4

0 MP λ−1/3 MP λ−1/4 MP ϕ

Figure 1.9: Evolution of the scalar field ϕ in the simplest field theory with the poten-
λ
tial V(ϕ) = ϕ4 , with quantum fluctuations of the field ϕ taken into account. When
4
4
ϕ> ∼ λ −1/4
MP (V(ϕ) >∼ MP ), quantum gravity fluctuations of the metric are large, and no
MP
classical description of space is possible in the simplest theories. For < ϕ < λ−1/4 MP ,
3 ∼ ∼
the field ϕ evolves relatively slowly, and the universe expands quasiexponentially. When
λ−1/6 MP < ∼ϕ≪λ
−1/4
MP , the amplitude of ϕ fluctuates markedly, leading to the endless
MP
birth of ever newer regions of the universe. For < ϕ ≪ λ−1/6 MP , fluctuations of the
3 ∼
field are of relatively low amplitude. The field ϕ rolls downhill, and fluctuations engender
MP
the density inhomogeneities required for the formation of galaxies. When ϕ < ∼ 3 , the
field starts to oscillate rapidly about the point ϕ = 0, particle pairs are produced, and all
of the energy of the oscillating field is converted into heat.

It is not hard to show that when ϕ ≪ ϕ∗ , where

ϕ∗ = λ−1/6 M , (1.8.4)

quantum fluctuations of ϕ have a negligible influence on its evolution, |δϕ(x)| ≪ ∆ϕ. It is

precisely at the later stages of inflation, when the field ϕ becomes less than ϕ∗ = λ−1/6 MP
that small inhomogeneities δϕ in the field ϕ and small density inhomogeneities δρ are
produced, leading to the formation of galaxies. On the other hand, when ϕ ≫ ϕ∗ , only
the mean field ϕ is governed by Eq. (1.7.22), and the role played by fluctuations becomes
extremely significant (see Fig. 1.8).
Consider a region of an inflationary universe of size ∆l ∼ H−1 (ϕ) that contains the
field ϕ ≫ ϕ∗ . According to the “no hair” theorem for de Sitter space, inflation in this
region of space proceeds independently of what happens in other regions. In such a
region, the field can be assumed to be largely homogeneous, as initial inhomogeneities of
the field ϕ are reduced by inflation, and nascent inhomogeneities (1.8.3) that make their
appearance during inflation have wavelengths l > H−1 . In the typical time ∆t = H−1 , the
 44

ϕ
δϕ
∆ϕ A
B

H -1 x
eH -1

Figure 1.10: Evolution of a field ϕ ≫ ϕ∗ = λ−1/6 MP in an inflationary region of the

universe of initial size ∆l = H−1 (ϕ). Initially (A), the field ϕ in this domain is relatively
homogeneous, since inhomogeneities δϕ(x) with a wavelength l ∼ H−1 (ϕ) that result from
H
inflation are of order δϕ ∼ ≪ ϕ. After a time ∆t = H−1 , the size of the region has
2π
grown (B) by a factor of e. When ϕ ≫ ϕ∗ , the average decrement ∆ϕ of the field ϕ
H
in this region is much less than |δϕ| ∼ . This means that in almost half the region
2π
under consideration, the field ϕ grows instead of shrinking. Thus, in a time ∆t = H−1 ,
e3
the volume occupied by increasing ϕ values grows by a factor of approximately ≈ 10.
2

region in question will have grown by a factor of e, and its volume will have grown by
a factor of e3 ≈ 20, so that it could be subdivided approximately into e3 regions of size
O(H−1 ), each once again containing an almost homogeneous field ϕ, which differs from
the original field ϕ by δϕ(x) − ∆ϕ ≈ δϕ(x). This means, however, that in something
e3
like regions of size O(H−1 ), instead of decreasing, the field ϕ increases by a quantity
3
H
of order |δϕ(x)| ∼ ≫ ∆ϕ (see Fig. 1.8). This process repeats during the next time
2π
−1
interval ∆t = H , and so on. It is not hard to show that the total volume of the universe
occupied by the continually growing field ϕ increases approximately as
√ ϕ2
!
exp[(3 − ln 2) H t] >
∼ exp 3 λ t ,
MP
while the total volume occupied by the non-decreasing field ϕ grows almost as fast as
exp[3 H(ϕ) t].
This means that regions of space containing a field ϕ continually spawn brand new
regions with even higher field values, and as ϕ grows the birth and expansion process in
 45

the new regions takes place at an ever increasing rate.

To better understand the physical meaning of this phenomenon, it is useful to examine
those regions which, while rare, are still constantly appearing, where the field ϕ increases
H(ϕ)
continuously, i.e., it is typically augmented by δϕ ∼ in each successive time interval
2π
−1
∆t = H (ϕ). The rate of growth of the field in such regions is given by

dϕ H2 (ϕ) 4 V(ϕ) λ ϕ4
= = = , (1.8.5)
dt 2π 3 M2P 3 M2P

whereupon
λt
ϕ−3 (t) = ϕ−3
0 − . (1.8.6)
M2P
This means that in a time
M2P
τ= (1.8.7)
λ ϕ30
the field ϕ should become infinite. Actually, of course, one can only say that the field in
such regions approaches the limiting value ϕ for which V(ϕ) ∼ M4P (i.e., ϕ ∼ λ−1/4 MP ).
At higher densities, it becomes impossible to treat such regions of space classically. Fur-
thermore, formal consideration of inflationary regions with V(ϕ) ≫ M4P indicates that
most of their field energy is concentrated not at V(ϕ), but at a value related to the
V2
inhomogeneities δϕ(x) and proportional to H4 ∼ 4 . Therefore, in the overwhelming
MP
majority of regions of the universe with V(ϕ) ≫ M4P , inflation is most likely cut short,
and in any case we cannot describe it in terms of classical space-time.
To summarize, then, many inflationary regions with V(ϕ) ∼ M4P are created over a
M2P
time τ ∼ in a part of the universe originally filled with a field ϕ0 ≫ ϕ∗ . Some
λ ϕ30
fraction of these regions will ultimately expand to become regions with V(ϕ) ≫ M4P ,
that is, a space-time foam, which we are presently not in a position to describe. It will
be important for us, however, that the volume of the universe filled by the extremely
large and non-decreasing field ϕ, such that V(ϕ) ∼ M4P , continue to grow at the highest
possible rate, as exp(c MP t), c = O(1). The net result is that in time t ≫ τ (1.8.7)
(in a synchronous coordinate system; see Section 10.3), most of the physical volume of
an original inflationary region of the universe with ϕ = ϕ0 ≫ ϕ∗ should contain an
exceedingly large field ϕ, with V(ϕ) ∼ M4P .
This does not at all mean that the whole universe must be in a state with the Planck
density. Fluctuations of the field ϕ constantly lead to the formation of regions not just
with ϕ ≫ ϕ∗ , but with ϕ ≪ ϕ∗ as well. Just such regions form gigantic homogeneous
regions of the universe like our own. After inflation, the typical size of each such region
exceeds
π(ϕ∗ )2
" #
4
−1
l ∼ MP exp 2
∼ M−1
P exp(π λ
−1/3
) ∼ 106·10 cm . (1.8.8)
MP
 46

7
when λ ∼ 10−14 . This is much less than l ∼ M−1 P exp(π λ
−1/2
) ∼ 1010 cm (1.7.40),
which we obtained by neglecting quantum fluctuations, but it is still hundreds of orders
of magnitude larger than the observable part of the universe.
A more detailed justification of the foregoing results [132, 133] has been obtained
within the context of a stochastic approach to the inflationary universe theory [134, 135]
(see Sections 10.2–10.4). We now examine two of the major consequences of these results.

1.8.1. The self-regenerating universe and the singularity problem

As we have already pointed out in the preceding section, the most natural initial value
of the field ϕ in an inflationary region of the universe is ϕ ∼ λ−1/4 MP ≫ ϕ∗ ∼ λ−1/6 MP .
Such a region endlessly produces ever newer regions of the inflationary universe containing
the field ϕ ≫ ϕ∗ . As a whole, therefore, this entire universe will never collapse, even if
it starts out as a closed Friedmann universe (see Fig. 1.8.1.). In other words, contrary
to conventional expectations, even in a closed (compact) universe there will never be a
global singular spacelike hypersurface — the universe as a whole will never just vanish into
nothingness. Similarly, there is no sufficient reason for assuming that such a hypersurface
ever existed in the past — that at some instant of time t = 0, the universe as a whole
suddenly appeared out of nowhere.
This of course does not mean that there are no singularities in an inflationary universe.
On the contrary, a considerable part of the physical volume of the universe is constantly
in a state that is close to singular, with energy density approaching the Planck density
V(ϕ) ∼ M4P . What is important, however, is that different regions of the universe pass
through a singular state at different times, so there is no unique end of time, after which
space and time disappear. It is also quite possible that there was no unique beginning of
time in the universe.
It is worth noting that the standard assertion about the occurrence of a general cos-
mological singularity (i.e., a global singular spacelike hypersurface in the universe, or
what is the same thing, a unique beginning or end of time for the universe as a whole)
is not a direct consequence of existing topological theorems on singularities in general
relativity [69, 70], or of the behavior of general solutions of the Einstein equations near
a singularity [68]. This assertion is primarily based on analyses of homogeneous cosmo-
logical models like the Friedmann or Bianchi models. Certain authors have emphasized
that there might in fact not be a unique beginning and end of time in the universe as a
whole if our own universe is only locally a Friedmann space but is globally inhomogeneous
(a so-called quasihomogeneous universe; see [34, 136]). However, in the absence of any
experimental basis for hypothesizing significant large-scale inhomogeneity of the universe,
this approach to settling the problem of an overall cosmological singularity has not elicited
much interest.
The present attitude toward this problem has changed considerably. Actually, the
only explanation that we are aware of for the homogeneity of the observable part of
the universe is the one provided by the inflationary universe scenario. But as we have
just shown, this same scenario implies that on the largest scales the universe must be
 47

Figure 1.11: A rather fanciful attempt to convey some impression of the global structure
of the inflationary universe. One region of the inflationary universe gives rise to a mul-
titude of new inflationary regions; in different regions, the properties of space-time and
elementary particle interactions may be utterly different. In this scenario, the evolution
of the universe as a whole has no end, and may have no beginning.

absolutely inhomogeneous, with density excursions ranging from ρ < ∼ 10

−29
g/cm3 (as in
4 94 3
the observable part of the universe) to ρ ∼ MP ∼ 10 g/cm . Hence, there is presently no
compelling basis for maintaining that there is a unique beginning or end of the universe
as a whole. (For a more detailed discussion of this question, see also Section 10.4.)
It is not impossible in principle that the universe as a whole might have been born
“out of nothing,” or that it might have appeared from a unique initial singularity. Such
a suggestion could be fairly reasonable if in the process a compact (closed, for example)
universe of size l = O(M−1P ) were produced. But for a noncompact universe, this hypoth-
esis is not only hard to interpret, it is completely implausible, since there would seem to
be absolutely no likelihood that all causally disconnected regions of an infinite universe
could spring simultaneously from a singularity (see the discussion of the horizon problem
in Section 1.5). Fortunately, this hypothesis turns out to be unnecessary for the scenario
being developed, and in that sense it seems possible to avoid one of the main conceptual
difficulties associated with the problem of the cosmological singularity [57].
 48

1.8.2. The problem of the uniqueness of the universe, and the

Anthropic Principle

The ceaseless creation of new regions of the inflationary universe takes place at the stage
when λ−1/6 MP < ∼ϕ< ∼λ
−1/4
MP ; then λ−1/3 M4P <
∼ V(ϕ) <
4
∼ MP (10 MP <
−5 4
∼ V(ϕ) <
4
∼ MP for
λ ∼ 10−14 ). In other words, it is not necessary to appeal for a description of this process
to speculative phenomena that take place above the Planck density.
On the other hand, it is important that much of the physical volume of the universe
must at all times be close to the Planck density, and expanding exponentially with a
Hubble constant H of order MP . In realistic elementary particle theories, apart from the
scalar field ϕ responsible for inflation, there are many other types of scalar fields Φ, H, etc.,
with masses m ≪ MP . Inflation leads to the generation of long-wave fluctuations not just
in the field ϕ, but in all scalar fields with m ≪ H ∼ MP . As a result, the universe becomes
filled with fields ϕ, Φ, and so forth, which vary slowly in space and take on all allowable
4
values for which V(ϕ, Φ, . . .) <
∼ MP . In those regions where inflation has ended, the scalar
fields “roll down” to the nearest minimum of the effective potential V(ϕ, Φ, . . .), and the
universe breaks up into exponentially large domains (mini-universes) filled with the fields
ϕ, Φ, etc., which in the different domains take on values corresponding to all the local
minima of V(ϕ, Φ, . . .). In Kaluza–Klein and superstring theories, quantum fluctuations
can result in a local change in the type of compactification on a scale O(H−1 ) ∼ O(M−1 P ). If
the region continues to inflate after this change, then by virtue of the “no hair” theorem
for de Sitter space, the properties of the universe outside this region (its size and the
type of compactification) cannot exert any influence on the region, and after inflation
an exponentially large mini-universe with altered compactification will have been created
[333].
The result is that the universe breaks up into mini-universes in which all possible types
of (metastable) vacuum states and all possible types of compactification that support infla-
tionary behavior are realized. We live in a region of the universe in which there are weak,
strong, and electromagnetic interactions, and in which space-time is four-dimensional. We
cannot rule out the possibility, however, that this is so not because our region is the only
one or the best one, but because such regions exist, are exponentially plentiful (or more
likely, infinitely plentiful), and life of our type would be impossible in any other kind of
region [57, 78].
This discussion is based on the Anthropic Principle, whose validity the author himself
previously cast into doubt (in Section 1.5). But now the situation is different — it
is not at all necessary for someone to sit down and create one universe after another
until he finally succeeds. Once the universe has come into being (or if it has existed
eternally), that universe itself will create exponentially large regions (mini-universes),
each having different elementary-particle and space-time properties. If good conditions
for the appearance of life in a solar-system environment are then to ensue, it turns out that
the same conditions necessarily appear on a scale much larger than the entire observable
part of the universe. In fact, for galaxies to arise in the simple model we are considering,
6·104
one must have λ ∼ 10−14 , and as we have seen, this leads to a characteristic size l > ∼ 10
 49

cm for the uniform region. Thus, within the context of the approach being developed, it
is possible to remove the main objections to the cosmological application of the Anthropic
Principle (or to be more precise, of the Weak Anthropic Principle; see Sections (10.5 and
10.7, where stronger versions of the Anthropic Principle are also discussed).
This result may have important methodological implications. Attempts to construct
a theory in which the observed state of the universe and the observed laws of interaction
between elementary particles are the only ones possible and are realized throughout the
entire universe become unnecessary. Instead, we are faced with the problem of construct-
ing theories that can produce large regions of the universe that resemble our own. The
question of the most reasonable initial conditions near a singularity and the probability
of an inflationary universe being created is supplanted by the question of what values the
physical fields might take, what the properties of space are in most of the inflationary
universe, and what the most likely way is to form a region of the universe of size Rp ∼ 1028
cm with observable properties and observers resembling our own.
This new statement of the problem greatly enhances our ability to construct realistic
models of the inflationary universe and realistic elementary particle theories.

1.9 Summary
In this introductory chapter, we have discussed some basic features of inflationary cos-
mology. One should take into account, however, that many details of the inflationary
universe scenario look different in the context of different theories of elementary particles.
For example, it is not necessary to assume that the field ϕ which drives inflation is an
elementary scalar field. In certain theories, the role played by this field can be assumed by
the curvature scalar R, a fermion condensate hψ̄ ψi or vector meson condensate hGaµν Gaµν i,
or even the logarithm of the radius of a compactified space. More detailed discussions
of phase transitions in the unified theories of weak, strong, and electromagnetic interac-
tions, of various versions of the inflationary universe scenario, and of different aspects of
quantum inflationary cosmology are to be found in subsequent sections of this book.
 2
Scalar Field, Effective Potential, and
Spontaneous Symmetry Breaking

2.1 Classical and quantum scalar fields

As we have seen, classical (or quasiclassical) scalar fields play an essential role in present-
day cosmological models (and also in modern elementary particle theories). We will
often deal with homogeneous and inhomogeneous classical fields, and there is sometimes
a question as to which fields can be considered classical, and in what sense.
Let us recall, first of all, that in accordance with the standard approach to quanti-
zation of the scalar field ϕ(x), the functions a+ (k) and a− (k) in (1.1.3) can be put into
correspondence with the creation and annihilation operators a+ −
k and ak for particles with
momentum k. The commutation relations take the form [58]
1
[ϕ− , ϕ+ ] ≡ [a− +
k , aq ] = δ(k − q) , (2.1.1)
2k0 k q
where the operator a−
k acting on the vacuum gives zero:

a−
k |0i = 0 ; h0| a+
k = 0 ; h0|ϕ(x)|0i . (2.1.2)

The operator a+
k creates a particle with momentum k,

a+
k |ψi = |ψ, ki , (2.1.3)

while the operator a−

k annihilates it.

a−
k |ψ, ki = |ψi . (2.1.4)

Now consider the Green’s function for the scalar field ϕ [58],

i e−ikx
Z
G(x) = h0|T[ϕ(x) ϕ(0)]|0i = d4 k . (2.1.5)
(2 π)4 m2 − k 2 − i ε
Here T is the time-ordering operator, and ε shows how to perform integration near the
singularity at k 2 = m2 (from here on, we omit both symbols). Evaluation of this expression
 51

indicates that when t = 0 and x > −1

∼ m , G(x) falls off exponentially with increasing x;
that is, the correlation between ϕ(x) and ϕ(0) becomes exponentially small. When m = 0,
G(x) has a power-law dependence on x.
It is also useful to calculate G(0), which after transforming to Euclidean space (by a
Wick rotation k0 → −i k4 ) may be written in the form

1 d4 k 1 d3 k
Z Z
G(0) = h0|ϕ2 |0i = = √ . (2.1.6)
(2 π)4 2
k +m 2 (2 π)3 2 k2 + m2
If averaging is carried out, for example, over a state containing particles rather than over
the conventional vacuum in Minkowski space, we can represent the quantity h0|ϕ2|0i ≡
hϕ2 i in the form

2 1 Z d3 k
hϕ i = √ (1 + 2ha+ −
k ak i)
(2 π)3 2 k2 + m2
1 d3 k 1
Z  
= √ + nk . (2.1.7)
(2 π)3 k2 + m2 2
Here nk is the number density of particles with momentum k. For instance, for a Bose
gas at nonzero temperature T, one has
1
nk = √  . (2.1.8)
exp k2 + m2 − 1
T

Another important example is a Bose condensate ϕ0 of noninteracting particles of the

field ϕ, with mass m and vanishing momentum k, for which

nk = (2 π)3 ϕ20 m δ(k) , (2.1.9)

or a coherent wave of particles with momentum p:

q
nk = (2 π)3 ϕ2p p2 + m2 δ(k − p) . (2.1.10)

In both cases, nk tends to infinity at some value of k. The fact that the a± k operators
of (2.1.1) do not commute can then be ignored, as nk ≫ 1 in (2.1.7). Therefore, the
condensate ϕ0 and the coherent wave ϕp can be called classical scalar fields. In performing
calculations, it is convenient to separate the field ϕ into a classical field (condensate) ϕ0
(ϕp ) and field excitations (scalar particles), with quantum effects being associated only
with the latter. This is formally equivalent to the appearance of a nonvanishing vacuum
average of the original field ϕ, h0|ϕ|0i = ϕ0 , and reversion to the standard formalism
(2.1.2) requires that we subtract the classical part ϕ0 from the field ϕ; see (1.1.12).
The foregoing instances are not the most general. If the condensate results from
dynamic effects (minimization of a relativistically invariant effective potential), the prop-
erties of its constituent particles will be altered, and the condensate itself (in contrast to
 52

(2.1.9) and (2.1.10)) can turn out to be relativistically invariant. This is precisely what
happens in theories incorporating the Glashow–Weinberg–Salam model, where hϕ2 i can
be put in the form

1 d3 k 1 d3 k
Z Z
2
hϕ i = √ + √ nk (2.1.11)
(2 π)3 2
2 k +m 2 (2 π)3 k2
√
with k = k2 , and
nk = (2 π)3 ϕ20 k δ(k) . (2.1.12)
The gist of this representation is that the constant classical scalar field ϕ0 (1.1.12) is
Lorentz invariant, and it can therefore only form a condensate if the particles comprising
it have zero momentum and zero energy — in other words, zero mass (compare (2.1.11)
and (2.1.7)).
It is not obligatory that the constant classical field be interpreted as a condensate,
but this proves to be a very useful, fruitful approach to the analysis of phase transitions
in gauge theories. There, the relativistically invariant form of the condensate (2.1.11),
(2.1.12) leads to a number of effects which are lacking from solid-state theory with a
condensate (2.1.9). We shall return
√ to this problem in the next chapter.
Note that nk ≫ 1 when k + m2 ≪ T for an ultrarelativistic Bose gas (2.1.8).
2

We
√ can therefore tentatively divide the field ϕ√into a quantum part corresponding to
k2 + m2 > 2 2
∼ T, and a (quasi)classical part with k +√m ≪ T. This sort of partitioning
is not very useful, however, since it is excitations with k2 + m2 ∼ T that make the main
contribution to most thermodynamic functions.
Much more interesting effects arise in the inflationary universe,
2
where the main contribution to hϕ i, to density inhomogeneities, and to a number of
other quantities comes precisely from long-wavelength modes with k ≪ H, for which
nk ≫ 1. The interpretation of these modes as inhomogeneous classical fields δϕ signifi-
cantly facilitates one’s understanding of a great many of the fundamental features of the
inflationary universe scenario. Corresponding effects were discussed in Section 1.8, and
we shall continue the discussion in Chapters 7 and 10.
Let us formulate a few more criteria that could help one decide whether the field ϕ
is (quasi)classical. One has already been discussed, namely the presence of modes with
nk ≫ 1. Another is the behavior of the correlation function of G(x) at large x. At large
x, this function usually (when there are no classical fields) falls off either exponentially or
according to a power law (as x−2 ). When there really is a condensate (2.1.9), (2.1.11) or
a coherent wave (2.1.10), the correlation function no longer decreases at large x (since the
condensate is everywhere the same, i.e., the values at different points are correlated). The
onset of long-range order is thus another criterion for the existence of a classical field in
a medium, one that has long been successfully applied in the theory of phase transitions.
As will be shown in Chapter 7, the corresponding correlation function in the inflationary
universe theory falls off only at exponentially large distances x ∼ H−1 exp(H t), H t ≫ 1,
which enables us to speak of a classical field δϕ(x) being produced during inflation.
 53

Figure 2.1: One- and two-loop diagrams for V(ϕ) in the theory of the scalar field (1.1.5).

Somewhat surprisingly, the classical field ϕ cannot be overly non-

uniform (unless it is a coherent wave with a single well-defined momentum (2.1.10)).
Suppose, in fact, that in some region of space ∇ϕ ∼ k ϕ ≫ m ϕ. In order for this field to
be distinguishable from the quantum fluctuation
q background, the field ϕ must be greater
2
than the contribution to the rms value hϕ i coming from quantum fluctuations with
momentum ∼ k ≫ m. Making use of (2.1.6), we obtain
ϕ2 > 2
∼ Ck , (2.1.13)
where C = O(1), or
(∇ϕ)2 <∼ϕ .
4
(2.1.14)
In particular, this means that the initial value of the classical scalar field ϕ cannot be
arbitrary; inhomogeneities in the classical scalar field cannot exceed a certain limit.
Even more important constraints can be obtained by taking quantum gravitation into
consideration. At energy densities of the order of the Planck density, fluctuations of the
metric become so large that one can no longer speak of classical space-time with a classical
metric gµν (in the same sense as one would speak of the classical field ϕ). This means
that it is impossible to treat fields ϕ as being classical unless
∂µ ϕ ∂ µ ϕ <
∼ M4P , µ = 0, 1, 2, 3 , (2.1.15)
V(ϕ) <
∼ M4P . (2.1.16)
We made essential use of these constraints in discussing initial conditions in the inflation-
ary universe in Section 1.7.

2.2 Quantum corrections to the effective potential V(ϕ)

In Section 1.1, we investigated the theory of symmetry breaking in the simplest quantum
field theoretical models, neglecting quantum corrections to the effective potential of the
 54

scalar field ϕ. Nevertheless, in some cases quantum corrections to V(ϕ) are substantial.
According to [137, 138], quantum corrections to the classical expression for the effective
potential are given by a set of all one-particle irreducible vacuum diagrams (diagrams
that do not dissociate into two when a single line is cut) in a theory with the Lagrangian
L(ϕ + ϕ0 ) without the terms linear in ϕ. Corresponding diagrams with one, two, or more
loops for the theory (1.1.5) have been drawn in Fig. 2.1. In the present case, expansion
in the number of loops corresponds to expansion in the small coupling constant λ. In the
one-loop approximation (taking only the first diagram of Fig. 2.1 into account),

µ2 2 λ 4 1
Z h i
V(ϕ) = − ϕ + ϕ + d4 k ln k 2 + m2 (ϕ) . (2.2.1)
2 4 2 (2 π)4

Here k 2 = k42 + k2 (i.e., we have carried out a Wick rotation k0 → −i k4 and integrated
over Euclidean momentum space), and the effective mass squared of the field ϕ is

m2 (ϕ) = 3 λ ϕ2 − µ2 . (2.2.2)

As before, we have omitted the subscript 0 from the classical field ϕ in Eqs. (2.2.1) and
(2.2.2). The integral in (2.2.1) diverges at large k. To supplement the definition given
by (2.2.1), it is necessary to renormalize the wave function, mass, coupling constant, and
vacuum energy [2, 8, 9]. To do so, we may add to L(ϕ + ϕ0 ) of (1.1.5) the counterterms
C1 ∂µ (ϕ + ϕ0 ) ∂ µ (ϕ + ϕ0 ), C2 (ϕ + ϕ0 )2 , C3 (ϕ + ϕ0 )4 and C4 .
The meaning of (2.2.1) becomes particularly clear after integrating over k4 . The result
(up to an infinite constant that is eliminated by renormalization of the vacuum energy,
i.e., by the addition of C4 to L(ϕ + ϕ0 )), is

µ2 2 λ 4 1 Z
3
q
V(ϕ) = − ϕ + ϕ + d k k 2 + m2 (ϕ) . (2.2.3)
2 4 2(2 π)3

Thus, in the one-loop approximation, the effective potential V(ϕ) is given by the sum of
the classical expression for the potential energy of the field ϕ and a ϕ-dependent vacuum
energy shift due to quantum fluctuations of the field ϕ. To determine the quantities Ci ,
normalization conditions must be imposed on the potential, and these, for example, can
be chosen to be [139]

dV 

= 0,
dϕ ϕ=µ/√λ
d2 V 

= 2 µ2 . (2.2.4)
dϕ2 ϕ=µ/√λ

These normalization√conditions are chosen to ensure that the location of the minimum
of V(ϕ) for ϕ = µ/ λ and the curvature of V(ϕ) at the minimum (which is the same
to lowest order in λ as the mass squared of the scalar field ϕ) remain the same as in
the classical theory. Other types of normalization conditions also exist; for example, the
 55

Figure 2.2: Diagrams for V(ϕ) in the Higgs model. The solid, dashed, and wavy lines
correspond to the χ1 , χ2 , and Aµ fields, respectively.

Coleman–Weinberg conditions [137] are

d2 V 

2
= m2 ,
dϕ ϕ=0 

d4 V 

= λ, (2.2.5)
dϕ4 ϕ=M
where M is some normalization point. All physical results obtained via the normaliza-
tion conditions (2.2.4) and (2.2.5) are equivalent, after one establishes the appropriate
correspondence between the parameters µ, m, M, and λ in the renormalized expressions
for V(ϕ) in the two cases. The conditions (2.2.4) are usually more convenient for practi-
cal purposes in work with theories that have spontaneous symmetry breaking, although
(2.2.5) is sometimes the most suitable approach in certain instances involving the study
of the fundamental features of the theory, since the first condition determines the mass
squared of the scalar field prior to symmetry breaking. Since we are most interested in
d2 V
the present section in the properties of V(ϕ) for certain values of m2 (ϕ) = at the
dϕ2
minimum of V(ϕ), we will use the conditions (2.2.4). The effective potential V (ϕ) then
takes the form [23]
µ2 (3 λ ϕ2 − µ2 )2 3 λ ϕ2 − µ 2
!
λ
V(ϕ) = − ϕ2 + ϕ4 + ln
2 4 64 π 2 2 µ2
21 λ µ2 2 27 λ2 4
+ ϕ − ϕ . (2.2.6)
64 π 2 128 π 2
Clearly, for λ ≪ 1, quantum corrections only become important for asymptotically
large ϕ (when λ ln(ϕ/µ) ≫ 1), where it becomes necessary to take account of all higher-
order corrections. When λ > 0, it becomes extremely difficult to sum all higher-order
corrections to the expression for V(ϕ) at large ϕ. This problem can only be solved for a
special class of λ ϕ4 theories discussed in the next section.
We can make much more progress in clarifying the role of quantum corrections in
theories with several different coupling constants. As an example, let us consider the
Higgs model (1.1.15) in the transverse gauge ∂µ Aµ . In the one-loop approximation, the
effective potential in that case is given by the diagrams in Fig. 2.2.
 56

d c
V

a
0
ϕ

3 e4 3 e4 3 e4
Figure 2.3: Effective potential in the Higgs model. a) λ > ; b) > λ > ;
16 π 2 16 π 2 32 π 2
3 e4
c) > λ > 0; d) λ = 0.
32 π 2

For e2 ≪ λ, the contribution of vector particles can be neglected, and the situation is
analogous to the one described above. When e2 ≫ λ, we can ignore the contribution of
scalar particles. In that event, the expression for V(ϕ) takes the form [139]
µ 2 ϕ2 3 e4 λ ϕ4 9 e4
! !
V(ϕ) = − 1− + 1−
2 16 π 2 λ 4 32 π 2 λ
4 4 2
!
3e ϕ λϕ
+ 2
ln . (2.2.7)
64 π µ2
3 e4
Clearly, then, when λ < , the effective potential acquires an additional minimum at
16 π 2
3 e4
ϕ = 0, and when λ < , this minimum becomes even deeper than the usual minimum
µ
32 π 2
at ϕ = ϕ0 = √λ ; see Fig. 2.2.
3 e4
Hence, when λ < , symmetry breaking in the Higgs model becomes energetically
16 π 2
ϕ
unfavorable. This effect is due not to large logarithmic factors like λ ln > 1, but to
µ ∼
special relations between λ and e2 (λ ∼ e4 ), whereby the classical terms in the expression
for the effective potential (2.2.7) become of the same order as the quantum corrections
to order e2 . Higher-order corrections to (2.2.7) are proportional to λ2 and e6 , and do not
lead to √
any substantial modification of the form of V(ϕ) given by (2.2.7), over the range
<
ϕ ∼ µ/ λ in which we are most interested.
 57

e2 µ2
We remark here that m2A = e2 ϕ20 = , m2ϕ = 2 λ ϕ20 up to higher-order corrections
λ
in e2 . This means that symmetry breaking is only favorable in the Higgs model if
3 e4 2
m2ϕ > m . (2.2.8)
16 π 2 A
The significance of this result for the Glashow–Weinberg–Salam model is that the mass of
the Higgs boson in that theory (more precisely, in the standard version with one kind of
Higgs boson and no superheavy fermions, and with sin2 θW ∼ 0.23) should be more than
approximately 7 GeV [139, 140],
mϕ >∼ 7 GeV . (2.2.9)
From Eq. (2.2.7), we also obtain bounds on the coupling constant between Higgs bosons,
1 d4 V 

λ (ϕ = ϕ0 ) = [139]. In fact, λ > 0, and
6 dϕ4 ϕ=ϕ0
e4
λ(ϕ0 ) = λ + , (2.2.10)
2 π2
and this means that V (ϕ) has a minimum at ϕ0 6= 0 if
11 e4
λ(ϕ0 ) > . (2.2.11)
16 π 2
and the minimum at ϕ = ϕ0 is deeper than the one at ϕ = 0 if
19 e4
λ(ϕ0 ) > , (2.2.12)
32 π 2
A bound like (2.2.12) in the Weinberg–Salam model yields
−3
λ(ϕ0 ) >
∼ 3 · 10 . (2.2.13)
With cosmological considerations taken into account, the corresponding bound can
be improved somewhat. As we have already said in the Introduction, symmetry was
2
restored in the early universe at T > ∼ 10 GeV in Glashow–Weinberg–Salam theory, and
the only minimum of V(ϕ, T) was the one at ϕ = 0. A minimum appears at ϕ 6= 0 only
as the universe cools, and if the effective potential then continues to have a minimum at
ϕ = 0, it is not clear a priori whether the field ϕ will be able to jump out of the local
minimum at ϕ = 0 to a global minimum at ϕ = ϕ0 ∼ 250 GeV, nor is it clear what
the properties of the universe would be after such a phase transition. By making use
of high-temperature tunneling theory [62], it has been shown that this transition has an
exceedingly low probability of occurrence in the Glashow–Weinberg–Salam model. The
phase transition can therefore only take place if the minimum of V(ϕ) at ϕ = 0 is very
d2 V 

shallow, ≪ µ2 . This then leads to a somewhat more rigorous bound on the mass
dϕ2 ϕ=0
of the Higgs boson [141–144],
mϕ >∼ 10 GeV . (2.2.14)
 58

ϕ0
0
ϕ

Figure 2.4: Effective potential in the theory (1.1.13) with mψ ≫ mϕ .

One particular case which is especially interesting from the standpoint of cosmology (as
d2 V 

well as from the standpoint of elementary particle theory) is that in which = 0.
dϕ2 ϕ=0
This is known as the Coleman–Weinberg theory [137]. The effective potential in this
theory, which is based on the Higgs model (1.1.15), takes the form

25 e4 ϕ4 ϕ40
!
4 ϕ
V(ϕ) = ϕ ln − + . (2.2.15)
128 π 2 ϕ0 4 4

25 e4 4
We have added the term ϕ here in order to ensure that V(ϕ0 ) = 0. In the SU(5)
512 π 2 0
model, the corresponding effective potential takes the form

25 g 4
!
ϕ 1 9
V(ϕ) = ln − + M4 , (2.2.16)
128 π 2 ϕ0 4 32 π 2 X

where g 2 is the SU(5) gauge coupling constant, MX is the mass of the X boson, and ϕ is
defined by Eq. (1.1.19). Equation (2.2.16) lay at the foundation of the first version of the
new inflationary universe scenario, so we will have a number of occasions to return to it.
Whereas quantum fluctuations of vector fields stimulate the dynamical restoration
of symmetry, quantum fluctuations of fermions enhance symmetry breaking. We now
consider the simplified σ-model (1.1.13) as an example. At large ϕ, the effective potential
in this theory is given by [145]

µ2 9 λ2 − 4 h4 4 λ ϕ2
!
λ
V(ϕ) = − ϕ2 + ϕ4 + ϕ ln . (2.2.17)
2 4 64 π 2 µ2

Fermions evidently make a negative contribution at large ϕ, and when 3 λ < 2 h2 , the
effective potential V(ϕ) is unbounded from below (Fig. 2.2).
 59

mψ , GeV
D

200
100
A

10

B C
0.1 1 10 100 1000 mϕ , GeV

Figure 2.5: The hatched region corresponds to allowable mass values for the Higgs boson,
4 1/4
X
mϕ , and heavy fermions, mψ (or more precisely, (mψi ) ) when one takes account of
i
both cosmological considerations and quantum corrections to the effective potential in the
Glashow–Weinberg–Salam model. The area bounded by the curve ABCD is the region of
µ
absolute phase stability with spontaneous symmetry breaking, ϕ = √ .
λ

Of course when ϕ → ∞, the one-loop approximation is longer applicable. However,

µ2 λ
if λ ≪ h2 , there is a range of values of the field ϕ (ϕ2 ∼ exp 4 ) for which V(ϕ) <
! λ h
µ
V √ , and the one-loop approximation still gives reliable results. Thus, in the σ-model
λ
µ
with λ ≪ h2 , or what is the same thing, with mϕ ≪ mψ , the state ϕ = √ is unstable,
λ
and strong dynamical symmetry breaking takes place.
We can readily generalize this result to a wider class of theories, including the Glashow–
Weinberg–Salam theory, which leads to a set of constraints on the mass of the Higgs meson
and the fermion masses in this theory [139–151]; see Fig. 2.2. We shall take advantage of
cosmological considerations in Chapter 6 to strengthen these constraints.

2.3 The 1/N expansion and the effective potential in the

λϕ4 /N theory
As a rule, it is not possible to study the behavior of the effective potential in standard
perturbation theory as ϕ → ∞, but theories that are asymptotically free in all coupling
constants constitute an important exception. For example, it can be shown that in a
massless λ ϕ4 theory with negative λ, V(ϕ) decreases without bound as ϕ → ∞ both in
 60

the classical approximation and when quantum corrections are taken into consideration
[137, 152]. It is difficult to use standard perturbation theory in λ to investigate the
behavior of V(ϕ) as ϕ → ∞ in the λ ϕ4 theory with λ > 0. There does exist a class of
theories, however, in which one can make substantial progress toward understanding the
properties of V(ϕ) for both small and large ϕ, bringing with it a number of surprising
results.
Consider the O(N) symmetric theory of the scalar field Φ = {Φ1 ,
. . . , ΦN }, with the Lagrangian

1 µ2 2 λ  2 2
L= (∂µ Φ)2 − Φ − Φ , (2.3.1)
2 2 4!N
√
where Φ2 = Φ2i . The field Φ may have a classical part Φ0 =
X
N {ϕ, 0, . . . , 0}. Let us
i
also introduce the composite field
λ 2
χ̂ = µ2 + Φ (2.3.2)
6N
with a classical part χ, and let us add to (2.3.1) the term
!2
3N λ 2
∆L = χ̂ − µ2 − Φ , (2.3.3)
2λ 6N

so that
1 3N 2 3N 2 1
L′ = L + ∆L = L = (∂µ Φ)2 − µ χ̂ + χ̂ − χ̂ Φ2 . (2.3.4)
2 λ λ 2
The theory described by (2.3.4) is equivalent to the theory (2.3.1), since the Lagrange
equation for the field χ̂ in the theory (2.3.4) is exactly (2.3.2), while the Lagrange equation
for the field Φ in the theory (2.3.4), taking (2.3.2) into account, gives the Lagrange
equation for the field Φ in the theory (2.3.1) [153]. In the one-loop approximation, the
effective potential V(ϕ, χ) ≡ N V(ϕ, χ) which corresponds to the theory (2.3.4) is given
by [154]
3 1 1 1
 
V(ϕ, χ) = − + 2
χ (χ − 2 µ2) + χ ϕ2
2 λ 96 π 2
χ2 χ
 
+ 2 ln 2 − 1 , (2.3.5)
128 π 2 M
where M is the normalization parameter, and

1 1 ϕ2 χ χ
 
2
(χ − µ ) + = + ln 2 . (2.3.6)
λ 96 π 2 6 96 π 2 M
The effective potential V(ϕ) ≡ N V(ϕ) in the original theory (2.3.1) is equal to V(ϕ, χ(ϕ)).
It is important to note that all of the higher-order corrections to Eqs. (2.3.5) and (2.3.6)
 61

Re V

I
V
0 _
II ϕ ϕ
V

Figure 2.6: Effective potential in the theory (2.3.1) with µ2 > 0.

contain higher powers of 1/N, and vanish in the limit as N → ∞. In that sense, Eqs.
(2.3.5) and (2.3.6) are exact in the limit N → ∞.
We now impose the following normalization conditions on µ2 and λ in (2.3.5) and
(2.3.6):

d2 V 

Re 2  = µ2 , (2.3.7)
dϕ ϕ=0
d4 V 

Re 4  = λ. (2.3.8)
dϕ ϕ=0

This then tells us that after renormalization, the parameter M2 in (2.3.5) should be put
equal to µ2 .
The signs of µ2 and λ in (2.3.7) and (2.3.8) are arbitrary. For simplicity, we will
consider the case in which µ2 > 0, λ > 0. The field χ is found to be a double-valued
function of ϕ when ϕ < ϕ̄, where
λ χ(ϕ̄)
1− 2
ln 2 = 0 . (2.3.9)
96 π µ
As a result, for ϕ < ϕ̄, the effective potential V(ϕ) turns out to be a double-valued
function of ϕ (with branches VI (ϕ) and VII (ϕ), VI > VII ; see Fig. 2.3) [154]. The
normalization conditions (2.3.7) and (2.3.8) hold on the upper branch of V(ϕ).
λ χ
On the branch VII , the field χ is extremely large ( 2
ln 2 > 1), and one may well
96 π µ
ask whether Eqs. (2.3.5) and (2.3.6) are actually valid for such large χ and for any large
but finite N. The answer to this question is affirmative, since on the branch VII , χ is large
in magnitude but finite, and is independent of N. For any arbitrarily large χ, there should
therefore exist an N such that corrections ∼ O(1/N) to Eqs. (2.3.5) and (2.3.6) for this χ
are small [155].
When ϕ = 0, as was shown in [153] to lowest order in 1/N, the Green’s function
Gχχ (k 2 ) of the field χ on the upper branch VI has a tachyon pole at k 2 = −µ2 e1/λ . Using
 62

Re λ

0 _
ϕ ϕ

Figure 2.7: Effective coupling constant λ(ϕ) in the theory (2.3.1).

the same arguments as above, it can be shown that higher-order corrections in 1/N to
Gχχ (k 2 ) can change the type of singularity at k 2 < 0, but they cannot alter the fact
that Gχχ (k 2 ) changes sign at k 2 < 0. Such behavior of Gχχ (k 2 ) is incompatible with the
Källén–Lehmann theorem, and indicates that the theory is unstable against production
of the classical field χ, the reason simply being that on the branch VI, the point ϕ = 0 is
not a minimum but a saddle point of the potential V(ϕ, χ), and a transition takes place
to the minimum at ϕ = 0 on the branch VII .
However, even this point is not an absolute minimum of V(ϕ). In fact, according to
(2.3.5) and (2.3.6),
 

ϕ4 iπ 
 
2

V(ϕ) = −4π 1 +  (2.3.10)
ϕ2  ϕ2 
ln 2 ln 2
 
µ µ
as ϕ → ∞. This means that the potential V(ϕ) is not bounded from below, and the
theory (2.3.1) is unstable against production of arbitrarily large fields ϕ [155].
A number of objections can be raised to this conclusion, the principal one being the
following. Equation (2.3.10) holds when N = ∞, but for any finite N there might exist a
field ϕ = ϕN so large that when ϕ > ϕN the expression (2.3.10) becomes unreliable; an
absolute minimum of V(ϕ) might thus exist for ϕ > ϕN .
One response to this objection can be found by combining the 1/N expansion and the
renormalization group equation [155]. In order to do so, we first note that the magnitude
d4 V
of the effective coupling constant λ(ϕ) = which can be calculated using (2.3.5) and
dϕ4
(2.3.6), behaves as shown in Fig. 2.3. This then leads to several consequences:
λ
a) for large enough N, a ϕ4 theory with λ > 0 is equivalent to a theory with λ < 0,
N
representing merely another branch of the same theory;
λ
b) contrary to the usual expectations, a ϕ4 theory with λ > 0 is unstable, while a
N
 63

theory with λ < 0 is metastable for small ϕ;

c) for large enough ϕ, Re λ becomes negative, and tends to zero with increasing ϕ.
The last of these is the decisive point. We may choose such a large value ϕ = ϕ1 that
λ is in fact small and negative, and such a large value of N(ϕ1 ) that the higher-order
corrections in powers of 1/N to the value of λ(ϕ) at ϕ ∼ ϕ1 are also small. We can
then make use of the renormalization group equation to continue the quantity λ(ϕ) from
λ
ϕ = ϕ1 to ϕ → ∞, since the ϕ4 theory is asymptotically free when λ < 0. We must
N
then integrate λ(ϕ) with respect to ϕ and obtain the value of V(ϕ). These calculations
result in a value for V(ϕ) identical to that obtained from (2.3.10), and thereby confirm
that the effective potential in this theory is actually unbounded from below for large ϕ
[155].
This conclusion turns out to be valid regardless of the sign of µ2 and λ at ϕ = 0.
Interestingly enough, spontaneous symmetry breaking, which ought to occur in the theory
(2.3.1) when µ2 < 0, actually takes place only on the upper (unstable) branch of V(ϕ); on
the lower (metastable) branch, the effective mass squared of the field ϕ is always positive,
and symmetry breaking does not occur [154].
These results are fairly surprising, and in many respects they are quite instructive.
Quantum corrections are found to lead to instability even in theories where this might
be least expected, such as (2.3.1) with µ2 > 0 and λ > 0. It turns out that for large
N, there is no spontaneous symmetry breaking in this theory when µ2 < 0; it has also
been found that in the theory (2.3.1), λ < 0 and λ > 0 actually represent two branches
of a single theory. These branches coalesce at exponentially large values of ϕ, and in
the limit of very large ϕ, the effective constant λ(ϕ) becomes negative and tends to zero
from below. The latter result, however, is not so very surprising, as that is just how
the effective constant λ ought to behave for large fields and large momenta, according to
a study based on the renormalization group equation (see [58], for example). This sort
of pathological behavior of the effective coupling constant λ also lay at the basis of the
so-called zero-charge problem [156, 157]. For a long time, the corresponding results were
viewed as being rather unreliable, and it was considered plausible that in many realistic
situations the zero-charge problem actually does not appear; for example, see [158]. On
the other hand, the principal objections to the reliability of the results presented in [156,
157] do not seem to apply to derivations based on the 1/N expansion [155, 159]. More
recently, the existence of the zero-charge problem in the theory λ ϕ4 has been sufficiently
well proven, both analytically [160] and numerically [161] (“triviality” of the theory λ ϕ4 ).
The foregoing analysis aids in an understanding of the essence of this problem using
the theory (2.3.1) as an example: according to our results, for large N, no theory of the
form (2.3.1) possesses both a stable vacuum and a nonvanishing coupling constant λ.
One question that then emerges is whether this result has any bearing on realistic
elementary particle theories with spontaneous symmetry breaking. First of all, then, let
us analyze just how serious the shortcomings of the theory (2.3.1) actually are. At first
glance, the presence of a pole at k 2 = −µ2 e1/λ on the upper branch of V(ϕ) does not seem
so terrible, since it is usually taken to mean that the low-energy physics does not “feel”
 64

the structure of the theory at superhigh momenta and masses. This is actually so at large
k 2 > 0. But the example of the theory (1.1.5) with symmetry breaking demonstrates that
the presence of a tachyon pole at k 2 = −µ2 < 0 leads to more rapid development of an
instability than would a large tachyon mass; see (1.1.6). The upper branch of the potential
V(ϕ) therefore actually corresponds to an unstable vacuum state (an analogous instability
also occurs in a multicomponent formulation of quantum electrodynamics at sufficiently
large N [159, 162]). On the other hand, when λ ≪ 1, the life-time of the universe at
the point ϕ = 0 on the lower branch turns out to be exponentially large, so the putative
instability of the vacuum in this theory in no way implies that it cannot correctly describe
1/λ
our universe. One possible problem here is that at a temperature T > ∼ µ e , the local
II
minimum at ϕ = 0 on the branch V also vanishes [155, 163], but in the inflationary
universe theory the temperature can never reach such high values.
Proceeding to a discussion of more realistic theories, it must be pointed our that when
λ ≪ 1, the tachyon pole on the upper branch of V(ϕ) is situated at |k 2 | ≫ M2P , and at the
point ϕ̄ where VI and VII merge, the effective potential V(ϕ) exceed the Planck energy
density M4P . In that event, as will be shown in the following section, all of the major
qualitative and quantitative results obtained neglecting quantum gravitation become un-
reliable. Furthermore, quantum corrections to V(ϕ) associated with the presence of other
matter fields can become important at lower momenta and densities. These corrections
will not change the form of V(ϕ) at small ϕ, but they can completely eliminate the insta-
bility that arises with large fields and momenta. Exactly the same thing happens with the
instability in the zero-charge problem when one makes the transition to asymptotically
free theories [3, 152].
The basic practical conclusion to be drawn from the last two sections is that for the
most reasonable relationship between the coupling constants (λ ∼ e2 ∼ h2 ≪ 1), quantum
corrections to V(ϕ) in theories of the weak, strong, and electromagnetic interactions
become important only for exponentially strong fields, so that the classical expression for
V(ϕ) is often a perfectly good approximation. Quantum corrections can often lead to
instability of the vacuum when the fields or momenta are exponentially large, but this
difficulty can in principle be avoided by a small modification of the theory without altering
the shape of the effective potential at small ϕ.

2.4 The effective potential and quantum gravitational effects

In our discussion of the inflationary universe scenario in Chapter 1, we often turned our
attention to fields ϕ ≫ MP . There is some question as to whether quantum gravitational
effects might substantially modify V(ϕ) under such conditions, ultimately invalidating the
chaotic inflation scenario. Such suspicions have been voiced by a number of authors (see
[164], for example), and we must therefore dwell specifically upon this question.
Gravitational corrections ∆V(ϕ) to the potential V(ϕ) are of a twofold nature. On
the one hand, they are associated with the gravitational interaction between vacuum
 65

Figure 2.8: Typical diagrams for V(ϕ) with gravitational effects taken into account. The
heavy lines correspond to the external classical field ϕ, the lighter lines to scalar particles
of ϕ, and the wavy lines to gravitons.

fluctuations, as in the Feynman diagrams shown in Fig. 2.4. The entire set of such
diagrams can be summed, with the final result being [165]
d2 V V Λ2 V2 (ϕ) Λ2
∆V(ϕ) = C1 · ln + C2 ln . (2.4.1)
dϕ2 M2P M2P M4P M2P
The Ci here are numerical coefficient of order unity, and Λ is the ultraviolet cutoff. These
corrections clearly diverge as Λ → ∞, and generally speaking, they fail to converge simply
to a renormalized version of the original potential V(ϕ). This is manifestation of the well
known difficulty associated with the nonrenormalizability of quantum gravitation. One
usually assumes, however, that at momenta of order MP , there should exist a natural cutoff
due either to the nontrivial structure of the gravitational vacuum, or to the fact that when
|k 2 | > 2
∼ MP , gravitation becomes part of a more general theory with no divergences. If, in
accordance with this assumption, Λ2 does not exceed M2P by many orders of magnitude,
then
d2 V V V2
∆V = C̃1 · + C̃2 , (2.4.2)
dϕ2 M2P M4P
where C̃i = O(1).
Note that, contrary to the often expressed belief, these corrections do not contain any
φn
terms of the type O(1) M n , which would make the theory ill-defined at φ > Mp . The main
P
reason why these terms are not generated by quantum gravity effects is that the field φ
by itself does not have any physical meaning. It enters the theory only via its effective
2
potential V and mass squared ddϕV2 , which is why the quantum gravity corrections have
the structure shown in Eq. (2.4.2).
It can readily be shown that when
d2 V
m2ϕ = 2
≪ M2P , (2.4.3)
dϕ
V(ϕ) ≪ M4P , (2.4.4)
 66

gravitational corrections to V(ϕ) are negligible. In particular, for the theory λ ϕ4, (2.4.4)
is a much stronger condition than (2.4.3); it holds when

ϕ ≪ ϕP = λ−1/4 MP . (2.4.5)

For λ ∼ 10−14 , we obtain a very weak bound on ϕ from (2.4.5):

ϕ ≪ 3000 MP . (2.4.6)

Thus, in a classical space-time in which (2.4.5) holds (see Section 1.7), the indicated
quantum gravitational corrections to V(ϕ) are negligible.
The other type of correction to V(ϕ) relates to the change in the spectrum of vacuum
fluctuations in an external gravitational field. However, inasmuch as the magnitude of
the field itself is proportional to V(ϕ), the corresponding corrections (for V(ϕ) ≪ M4P )
are usually negligible. The most important exception to this is the contribution to V(ϕ)
from long-wavelength fluctuations of the scalar field ϕ that are generated at the time of
inflation. But as we already noted in Section 1.8, taking this effect into account does
not lead to any problems with the realization of the chaotic inflation scenario; in fact, on
the contrary, it engenders a self-sustaining inflationary regime over most of the physical
volume of the universe. We shall return to this question in Chapter 10.
 3
Restoration of Symmetry at High
Temperature

3.1 Phase transitions in the simplest models with spontaneous symmetry

breaking
Having discussed the basic features of spontaneous symmetry breaking in quantum field
theory, we can now turn to a consideration of symmetry behavior in systems of particles
in thermodynamic equilibrium which interact in accord with unified theories of the weak,
strong, and electromagnetic interactions. We will primarily examine systems of scalar
particles ϕ with the Lagrangian (1.1.5). Such particles carry no conserved charge, nor
is their number a conserved quantity. The chemical potential therefore vanishes for such
particles, and their density in momentum space is
1
nk = ! , (3.1.1)
k0
exp −1
T
√
where k0 = k2 + m2 is the energy of a particle with momentum k and mass m. All
particles disappear at T = 0 (nk → 0), and we revert to the situation described in the
previous chapter.
At finite temperature, all physically interesting quantities (thermodynamic potentials,
Green’s functions, etc.) in this system are given not by vacuum averages, but by Gibbs
averages
H
   
Tr exp − ...
h. . .i = T  (3.1.2)
H
 
Tr exp −
T
where H is the system Hamiltonian. In particular, the symmetry breaking parameter (the
“classical” scalar field ϕ) in this system is given by ϕ(T) = hϕi, rather than by h0|ϕ|0i.
In order to investigate the behavior of ϕ(T) at T 6= 0, let us consider the Lagrange
equation for the field ϕ in the theory (1.1.5),

( + µ 2 − λ ϕ2 ) ϕ = 0 , (3.1.3)
 68

and let us take the Gibbs average of this equation, giving

ϕ(T) − [λ ϕ2 (T) − µ2 ] ϕ(T) − 3 λ ϕ(T) hϕ2i − λ hϕ3 i = 0 . (3.1.4)

Here, as in the analysis of spontaneous symmetry breaking at T = 0, we have separated

out the analog of the classical field ϕ by carrying out the shift ϕ → ϕ + ϕ(T), such that

hϕi = 0 . (3.1.5)

To lowest order in λ, hϕ3 i is equal to zero, whereas

1 d3 k
Z
hϕ2 i = 3
√ (1 + 2 ha+ −
k ak i)
(2 π) 2 k +m2 2

1 d3 k 1
Z  
= √ + nk . (3.1.6)
(2 π)3 k 2 + m2 2
The first term in (3.1.6) vanishes after renormalizing the mass of the field ϕ in the field
theory (at T = 0). As a result,

hϕ2 i = F(T, m(ϕ))

1 Z∞ k 2 dk
= q  .
2 π2 0 q

k 2 + m2 (ϕ)
k 2 + m2 (ϕ) exp − 1
T
(3.1.7)

Clearly, all interesting effects in this theory (λ ≪ 1) take place at T ≫ m, where we can
neglect m in (3.1.7). Then
T2
hϕ2 i = F(T, 0) = , (3.1.8)
12
and Eq. (3.1.4) becomes
" #
λ
ϕ(T) − λ ϕ (T) − µ + T2 ϕ(T) = 0 .
2 2
(3.1.9)
4

From (3.1.9), we obtain for the constant field ϕ(T)

" #
λ
ϕ(T) λ ϕ (T) − µ + T2 = 0 .
2 2
(3.1.10)
4

At sufficiently low temperature, this equation has two solutions,

1) ϕ(T) = 0 ;
s
µ2 T2
2) ϕ(T) = − . (3.1.11)
λ 4
 69

ϕ m2

0 0
Tc T Tc T

Figure 3.1: The quantities ϕ(T) and m2 (T) in the theory (1.1.5). The dashed lines
correspond to the unstable phase ϕ = 0 at T < Tc .

The second of these vanishes above a critical temperature

2µ
Tc = √ = 2 ϕ0 . (3.1.12)
λ
To derive the excitation spectrum at T 6= 0, we must carry out the shift ϕ → ϕ + δϕ
in (3.1.9). When ϕ(T) = 0, the corresponding equation takes the form
!
λ
δϕ − µ + T2
2
δϕ = 0 , (3.1.13)
4

which corresponds to a mass

λ 2
m2 = −µ2 + T (3.1.14)
4
for the scalar field at ϕ = 0. This quantity is negative when T < Tc , and s it becomes
µ2 T2
positive when T > Tc . For the solution of the second of Eqs. (3.1.11), ϕ(T) = − ,
λ 4
it takes the value
λ
m2 = 3 λ ϕ2(T) − µ2 + T2 = 2 λ ϕ2(T) . (3.1.15)
4
This solution is therefore stable for T < Tc , and it vanishes for T > Tc at the instant
when the solution ϕ = 0 becomes stable. This then means that a phase transition with
restoration of symmetry takes place at a temperature T = Tc [18–24].
We illustrate the foregoing results in Fig. 3.1. The quantity clearly decreases smoothly
with increasing temperature, corresponding to a second-order phase transition.
These results can also be obtained in a different way, based on a finite-temperature
generalization of the concept of the effective potential V(ϕ). We will not dwell on this
problem, noting simply that at its extremuma, the effective potential V(ϕ, T) coincides
with the free energy F(ϕ, T). To calculate V(ϕ, T), it suffices to recall that at T 6= 0,
quantum statistics is equivalent to Euclidean quantum field theory in a space which is
 70

periodic, with period 1/T along the “imaginary time” axis [166, 20]. To go from V(ϕ, 0)
to V(ϕ, T) one should replace all boson momenta k4 in the Euclidean integrals by 2 π n T
for bosons and (2 n + 1) π T for fermions, and sum over n instead of integrating over k4 :
Z ∞
X
dk4 → 2 π T . For example, at T 6= 0, Eq. (2.2.1) for V(ϕ) in the theory (1.1.5)
n=−∞
transforms into
µ2 2 λ 4
V(ϕ, T) = − ϕ + ϕ
2 4
∞ Z
T
d3 k ln[(2 π n T)2 + k 2 + m2 (ϕ)] ,
X
+ 3
2 (2 π) n=−∞
(3.1.16)

where m2 (ϕ) = 3 λ ϕ2 − µ2 . This expression can be renormalized using the same countert-
erms as for T = 0. Equation (1.2.3) gives the result of calculating V(ϕ, T) for T ≫ m. It
dV
is straightforward to show that the equation = 0, which determines the equilibrium
dϕ
d2 V
values of ϕ(T), is the same as (3.1.10), and that the quantity , which determines the
dϕ2
mass squared of the field ϕ, coincides with (3.1.14) and (3.1.15) (for equilibrium ϕ(T)).
The description of the phase transition in terms of the behavior of V(ϕ, T) is given in
Section 1.2.
The methods developed above can readily be generalized to more complicated models.
In the Higgs model (1.1.15), for example, in the transverse gauge ∂µ Aµ = 0, we have
* +
δL
= ϕ(T) [µ2 − λ ϕ2(T) − 3 λ hχ21 i − λ hχ22 i + e2 hA2µ i] = 0 (3.1.17)
δϕ

instead of Eq. (3.1.4). To start with, let us assume that λ ∼ e2 . Then the phase transition
takes place at T ≫ mχ , mA , as in the theory (1.1.5); hence

1 T2
hχ21 i = hχ22 i = − hA2µ i = , (3.1.18)
3 12
and Eq. (3.1.17) becomes

4 λ + 3 e2 2
!
2 2
ϕ λϕ − µ + T =0. (3.1.19)
12

This then implies that the phase transition takes place in the Higgs model at a critical
temperature
12 µ2
T2c1 = . (3.1.20)
4 λ + 3 e2
According to (3.1.19), ϕ(T) is a continuous function of T, that is, this is a second-order
phase transition [18–20].
 71

ϕο ϕ1

A
B
ϕ2

0 Tc1 Tc Tc 2 T
3 e4 4
Figure 3.2: The function ϕ(T) in the Higgs model with <λ<∼ e . The heavy curve
16 π 2
corresponds to the stable state of the system. Arrows indicate the behavior of ϕ with
increasing (A) and decreasing (B) temperature.

4 eµ
If we consider the case λ <∼ e , however, we find that mA (Tc1 ) ≈ λ > ∼ Tc1 , i.e., we
no longer have T ≫ mA , and the contribution of vector particles to (3.1.19) at T ∼ Tc1 is
strongly suppressed. We can then no longer neglect mA compared with T when calculating
hA2µ i = −F(T, mA ), and all of the equations are significantly altered. The simplest way to
understand this is to note that when m < T, the quantity F(T, mA ) can be represented
m
by a power series in :
T
T2 m2
" !#
3 m
F(T, m) = 1− +O . (3.1.21)
12 π T T2
Bearing in mind, then, that in the lowest order of perturbation theory mA = e ϕ, Eq.
(3.1.19) can be rewritten as
4 λ + 3 e2 2 3 e3
!
ϕ λ ϕ2 − µ 2 + T − Tϕ = 0 . (3.1.22)
12 4π
In contrast to (3.1.19), this equation has three solutions rather than two in a certain
temperature range Tc1 < T < Tc2 , corresponding to three different extrema of V(ϕ, T);
see Fig. 3.1. The solution ϕ = 0 is metastable when T > Tc1 . Upon heating, a phase
transition from the phase ϕ = ϕ1 to the phase ϕ = 0 begins at a temperature Tc , where
V(ϕ1 (Tc ), Tc ) = V(0, Tc ) . (3.1.23)
4
In the Higgs model with λ <
∼e,
!1/4
15 λ
Tc = µ; (3.1.24)
2 π2
 72

ϕο

0 Tc Tc 2 T
3 e4
Figure 3.3: The function ϕ(T) in the Higgs model with λ < .
16 π 2

see [23]. Clearly, the phase transition in the present case is a discontinuous one — a
first-order phase transition (see Fig. 3.1).
3 e4
Recall now that when λ < ∼ 16 π 2 , quantum corrections to V(ϕ, T) lead to the existence
3 e4
of a local minimum of V(ϕ) even at T = 0 (see Fig. 3.1), and when λ < , this
32 π 2
µ
minimum becomes deeper than the usual one at ϕ = √ ; see Section 2.2. Thus, as
λ
3 e4
λ→ the critical temperature Tc → 0. This does not mean, however, that a phase
32 π 2
transition in such a theory becomes easy to produce in a laboratory. The point here is that
a first-order phase transition occurs by virtue of the sub-barrier creation and subsequent
growth of bubbles of the new phase. Bubble formation is often strongly suppressed, so it
may take an extremely long time for the phase transition to occur. When the system is
heated, the phase transition therefore really takes place from a superheated phase ϕ1 at
some temperature higher than Tc . Likewise, when the system is cooled, a first-order phase
transition takes place from a supercooled phase at T < Tc . We will consider the theory of
bubble production in Chapter 5, and the cosmological consequences of first-order phase
transitions will be discussed in Chapters 6 and 7.

3.2 Phase transitions in realistic theories of the weak, strong, and electro-
magnetic interactions
As we have shown in the preceding section, when the coupling constants λ and e2 are
related in the most natural way, the phase transition in the Higgs model is second-order,
but when λ ∼ e4 , it becomes first-order. One can readily show that the same is true of a
 73

phase transition in the Weinberg–Salam theory. This implies that the phase transition in
the Weinberg-Salam model is first order only if the Higgs mass is sufficiently small.
14
On the other hand, the transitions that take place in grand unified theories at T >∼ 10
GeV, as a rule, prove to be first-order transitions with a considerable jump in the field ϕ
at the critical point [104]. There are two reasons why this is so. First, at T ∼ 1014 GeV,
the effective gauge constant g 2 ∼ 0.3; that is, it is three times the value of e2 at T ∼ 102
GeV. Second, there are a great many particles in grand unified theories that contribute
to temperature corrections to the effective potential. The net result is that the critical
temperature Tc1 of the phase transition turns out to be approximately of the same order
of magnitude as the particle masses at that temperature. As we showed in Section 3.1,
this is precisely the circumstance that leads to a first-order phase transition.
As an example, consider a theory with SU(5) symmetry [91]. The effective potential
in the simplest version of this theory is

µ2 a b
V(Φ) = − Tr Φ2 + (Tr Φ2 )2 + Tr Φ4 , (3.2.1)
2 4 2
7
where Φ is a traceless 5 × 5 matrix. If one takes b > 0, λ > 0, where λ = a + b,
15
the symmetric state Φ = 0 is unstable with respect to the appearance of the scalar field
(1.1.19), s
2 3 3
 
Φ= ϕ · diag 1, 1, 1, − , − , (3.2.2)
15 2 2
which breaks the SU(5) symmetry to SU(3) × SU(2) × U(1). At T = 0, the minimum of
µ
V(ϕ) corresponds to ϕ0 = √ . Temperature corrections to V(ϕ) come from 24 different
λ
kinds of Higgs bosons and 12 X and Y vector bosons. As a result, the counterpart of Eq.
(??) for ϕ(T) in the SU(5) theory is of the form [104]

Tϕ
 
2 2 2
ϕ µ −β T −λϕ − Q(g 2 , λ, b) = 0 , (3.2.3)
30 π
where
75 g 2 + 130 a + 94 b
β = , (3.2.4)
60
√ 16 √ √
Q = 7 λ 10 b + b 10 b + 3 15 λ3/2
3
√ 75 √ 3
+ 2 15 λ g + 2g . (3.2.5)
4
As we have already noted, a phase transition with symmetry breaking upon cooling takes
place somewhere in the temperature range between Tc1 and Tc , where Tc1 is given by
µ
Tc1 = √ . (3.2.6)
β
 74

To estimate the size of the jump at the phase transition point, let us determine the
quantity ϕ1 (Tc1 ), Fig. 3.1. At T = Tc1 , the first two terms in (3.2.3) cancel, and we find
that
Q ϕ0
ϕ1 (Tc1 ) = √ . (3.2.7)
30 π β λ
For the most reasonable values of the parameters a ∼ b ∼ g 2 = 0.3, Eqs. (3.2.4)–(3.2.7)
imply that
ϕ1 (Tc1 ) ∼ 0.75 ϕ0 , (3.2.8)
i.e., the jump in the field at the time of the phase transition is very large (of the same
order of magnitude as ϕ0 ).
In the discussion above, we studied only one “channel” of the phase transition, in
which the transition goes directly from the SU(5) phase to the SU(3) × SU(2) × U(1)
phase. In actuality, the phase transition usually entails the formation of an SU(4) × U(1)
intermediate phase, and other phases as well [67, 42]. Each of the intermediate phase
transitions is also a first-order transition. The kinetics of the phase transition in the
minimal SU(5) theory will be discussed in Chapter 6.

3.3 Higher-order perturbation theory and the infrared

problem in the thermodynamics of gauge fields
Our analysis of the high-temperature restoration of symmetry in the theory (1.1.5) (Sec-
tion 3.1) was based on the use of the lowest-order perturbation theory in λ. One may
then ask how reliable the results obtained in this manner really are.
This is not a completely trivial question. For example, apart from small terms ∼ λn T4 ,
λn T2 m2 , high-order corrections in λ to the expression for V(ϕ, T) at T 6= 0 could contain
terms proportional to m−n . Such terms become large when m is small.
In order to analyze this question more thoroughly, let us examine the N-th order
diagrams in λ for V(ϕ, T) in the theory (1.1.5), for ϕ = 0. The contribution of these
diagrams to V(0, T) can be written out as an expression of the form
Z
VN (0, T) ∼ (2 π T)N+1 λN d3 p1 . . . d3 pN+1
∞ 2N
[(2 π rk T)2 + q2k + m2 (T)]−1 ,
X Y
× (3.3.1)
ni =−∞ k=1

where qk is a homogeneous linear combination of the pi , and rk is a corresponding com-

bination of the ni , i = 1, . . . , N + 1, k = 1, . . . , 2N. When m → 0, the leading term in
the sum over ni is the one for which all ni = 0 (rk = 0), since the factors containing the
terms (2 π rk T)2 are nonsingular as m → 0, qk → 0. This leading term is given by
Z 2N
N+1 N
d3 p1 . . . d3 pN+1 [q2k + m2 (T)]−1
Y
∆VN (0, T) ∼ (2 π T) λ
k=1
 75

!N−3
3 4 λT
∼ λ T . (3.3.2)
m(T)
!N−3
λT
It can be seen that dangerous terms ∼ appear in the expression for V(0, T)
m
starting with perturbations of order N = 4, and these make it impossible to obtain reliable
results from perturbation theory with m < λ T. Fortunately, however, (3.1.14) can be used
to show that m ≫ λ T everywhere outside a small region near the critical temperature
Tc , within which
|T − Tc | <
∼ λ Tc . (3.3.3)
Everywhere outside this region, the results obtained in the preceding two sections are
reliable.
Matters are much more difficult when it comes to dealing with phase transitions in
the non-Abelian gauge theories that describe the interaction of Yang–Mills fields Aaµ with
one another and with scalar fields ϕ with a coupling constant g 2. At T 6= 0 in such
theories, higher-order perturbation terms ∼ g 2N grow with increasing N (as N → ∞)
2
for mA < ∼ g T. Since in the classical approximation, mA goes to zero (mA ∼ g ϕ(T))
at all temperatures above the critical value T = Tc , we are left with the question of
whether high-temperature corrections lead to sufficiently high mass mA (T) 6= 0, with a
!N
g2 T
corresponding cutoff of infrared-divergent powers of the form .
mA
The authors of [168, 169] have shown that high-temperature effects give rise to a pole
of the Green’s function of the Yang–Mills fields Gab
µν (k) at k0 ∼ g T, k = 0. This might be
interpreted as the appearance of an infrared cutoff at a mass mA ∼ g T, making the terms
!N
g2 T
small; actually, however, that would not be correct. The foregoing analysis tells
mA
us that the leading infrared divergences as mA → 0 are associated not with the behavior
of the Green’s functions at k = 0, k0 6= 0, but with their behavior when k0 = 0, k → 0
(k0 = 0 corresponds to ni = 0 in (3.3.1)). In this limit, the behavior of Gab µν (k) is most
easily studied in the Coulomb gauge, for which [166, 24]

Gab ab 2
00 = δ [k + π00 (k)]
−1
, (3.3.4)
ab ab
Gi0 = G0j = 0 , (3.3.5)
!
ki kj
Gab
ij = δ ab
δij − 2 G(k) , (3.3.6)
k

where k = |k|, a and b are isotopic spin indices, π00 (0) ∼ g 2 T2 to lowest order g 2 , and i,
j = 1, 2, 3.
Thus, there really is an infrared cutoff at m0 ∼ g T in Gab 00 , corresponding to the
usual Debye screening of the electromagnetic field in hot plasma [166]. A well known
result in quantum electrodynamics, however, is that a static magnetic field in plasma
cannot be screened, and there is consequently no infrared cutoff in Gij (k = 0, k → ∞)
 76

to any order of perturbation theory [166]. In the Yang–Mills gas, there will likewise be
no infrared cutoff at k0 = 0, k → 0 with momentum k ∼ g T. There may in principle
be one, however, at momentum k ∼ g 2 T, inasmuch as massless Yang–Mills particles (in
contrast to photons) interact with each other directly, and the same infrared divergences
appear in the thermodynamics of a Yang–Mills gas as in scalar field theory at the point
of a second-order phase transition. The difference is that the mass of a scalar field at
a phase transition point vanishes “by definition” (the curvature of V(ϕ) changes sign
at the phase transition point), while in the thermodynamics of a Yang–Mills gas, the
presence or absence of an infrared cutoff does not follow from any general considerations,
which imply only that the expected scale of the infrared cutoff is k ∼ g 2 T. One can
reach the same conclusion by analyzing the most strongly infrared-divergent part of the
theory [170], as well as the specific diagrams that could contribute to such a cutoff [24,
171, 172]. Unfortunately, when k ∼ g 2 T, all high-order corrections to the diagrams for
the polarization operator of the Yang–Mills field are of comparable size, so the infrared
2
behavior of the Green’s functions of the Yang–Mills field at k < ∼ g T thus far remains an
open problem. Meanwhile, the degree of confidence that we have in our understanding
of many of the fundamental features of gauge-theory thermodynamics depends on the
solution to this problem. Let us consider the three main possibilities, which illustrate the
significance of this problem.
1) There is no infrared cutoff at k ∼ g 2 T in the thermodynamics of a Yang–Mills
gas. In that case, higher-order perturbations will be larger than lower-order ones for
all thermodynamic quantities, making it impossible to use perturbation theory to study
the thermodynamic properties of gauge theories at T > Tc . The only thing we might
possibly be able to verify with reasonable assurance is that the energy density should
be proportional to T4 at superhigh temperature (from dimensional considerations). This
would only suffice for the crudest approach to the theory of the evolution of a hot universe
at T > Tc .
2) There is a tachyon pole or a sign change at some momentum k ∼ g 2 T in the Green’s
function Gab ij (k). This implies instability with respect to creation of classical Yang–Mills
fields. The second case is particularly interesting — the instability could result in the
spontaneous crystallization of the Yang–Mills gas at superhigh temperature, which might
lead to nontrivial cosmological consequences.
3) In the best possible case (from the standpoint of perturbation theory), the theory
contains a cutoff by virtue of the fact that G−1/2 (0) turns out to be a positive quantity
m(T) of order g 2 T. In that event, one can reliably calculate several of the lowest-order
perturbation terms in g 2 for the thermodynamic potential of the Yang–Mills gas (up to
∼ g 6 T4 ) [171, 172]. In principle, the appearance of such a cutoff can lead to monopole
confinement in a hot Yang–Mills plasma [173]; see Chapter 6.
Thus, the thermodynamic properties of hot dense matter described by gauge theories
are still far from being well-understood, and one must not lose sight of the inherent
difficulties and uncertainties. Nevertheless, many results have been reasonably reliably
established. As applied to the theory of phase transitions studied in this chapter, the
infrared problem in the thermodynamics of a Yang–Mills gas does not modify the results
 77

obtained for T < Tc . It can also be shown that ϕ(T) should be much smaller than
ϕ0 when T > Tc , and that at large T it cannot exceed O(g T). (If one assumes that
ϕ(T) ≫ g T, then the Yang–Mills fields acquire a mass mA ≫ g 2 T, perturbation theory
becomes reliable, and the latter predicts that ϕ(T) = 0 at T > Tc .) One should keep
these uncertainties in mind in discussing such complicated problems as production and
evolution of monopoles in grand unified theories. However, for most of the effects to be
discussed below, these uncertainties will not be important, and we will usually presume
that ϕ(T) at a sufficiently high temperature T > Tc , in accordance with the results
obtained in the preceding sections.
We must now state one last (but very important) reservation. We have assumed
throughout that the field ϕ has sufficient time to roll down to a minimum of V(ϕ, T).
This natural assumption is valid if the field ϕ is not too large initially — violations of this
“rule” are just what lead to the chaotic inflation scenario, which we discussed in Chapter
1 and will examine further in Chapter 7.
 4
Phase Transitions in Cold Superdense
Matter

4.1 Restoration of symmetry in theories with no neutral

currents
In Chapters 2 and 3, we studied phase transitions in hot superdense matter, where the
increase in density resulted from an increase in temperature. But it is also possible to
examine phase transitions in cold superdense matter, where the density is increased by
increasing the density of conserved charge or the number of particles at zero temperature
T. In the first papers in which this problem was studied it was claimed that raising the
density of cold matter would also result in the restoration of symmetry [25, 26]. The
basic idea in those papers was that the energy of fermions interacting with a scalar field
is proportional to g ϕ hψ̄ ψi. When the fermion density j0 = hψ̄ γ0 ψi increases, so does
hψ̄ ψi, and states with ϕ 6= 0 become energetically unfavorable.
As an example, consider the theory (1.1.13) with the Lagrangian

1 µ2 2 λ 4
L= (∂µ ϕ)2 + ϕ − ϕ + ψ̄ (i ∂µ γµ − h ϕ) ψ . (4.1.1)
2 2 4
It is possible for fermions with j0 6= 0 to exist if they have a chemical potential α, which
(for α ≫ mψ = h ϕ) is related to j0 [61] by

α3
j0 = . (4.1.2)
3 π2
To calculate the corrections to V(ϕ) which appear due to the existence of the current j0
(4.1.2), we must add i α to the component k4 of the momentum of the fermions when the
one-loop contribution of fermions to V(ϕ) is computed [166]. As a result, the equation
for the equilibrium value of the field ϕ in the theory (4.1.1) takes the form [25, 26]
 !1/3 
dV h2 g j02
= 0 = ϕ  λ ϕ2 − µ 2 +  . (4.1.3)
dϕ 2 π2
 79

µ
When j0 = 0, we obtain ϕ = ± √ , as before. But it is clear that the presence of fermions
λ
with j0 6= 0 changes the effective value of µ2 , and for j0 > jc , where
√  
2π 2 µ 3
jc = , (4.1.4)
3 h
symmetry is restored in the theory (4.1.1).

4.2 Enhancement of symmetry breaking and the

condensation of vector mesons in theories with
neutral currents
Effects leading to the restoration of symmetry in the theory (4.1.1) appear only due to
quantum corrections to V(ϕ) for α 6= 0. This is because the fermion current jµ = hψ̄ γµ ψi
in the model (4.1.1) does not interact directly with any physical fields. At the same
time [27, 24], for realistic theories with neutral currents, in which the fermion current jµ
interacts with the neutral massive vector field Zµ , an increase in the fermion density j0
leads to an enhancement of symmetry breaking, while the effects considered in [25, 26] are
but minor quantum corrections relative to the effects examined in [27, 24]. Subsequent
study of this problem has shown that the effects appearing in cold superdense matter do
not simply amount to enhanced symmetry breaking. At high enough density, a condensate
of charged vector fields appears, and a redistribution of charge takes place among bosons
and fermions [28, 29].
As an example, let us consider the effects that take place in the Glashow–Weinberg–
Salam theory with λ ≫ e4 in the presence of a nonvanishing neutrino density nν =
1
hν̄e γ0 (1 − γ5 ) νe i. The conserved fermion density in this theory is
2
1
l = ē γ0 e + ν̄e γ0 (1 − γ5 ) νe . (4.2.1)
2
Clearly, given a lepton charge density hli, the most energetically favorable fermion distri-
bution is
1 1
neR = hē γ0 (1 + γ5 ) ei = neR hē γ0 (1 − γ5 ) ei = nν .
2 2
This would imply the appearance of a large charge density of electrons, however, which
is only possible if some sort of charge-cancelling subsystem comes into being at the same
time. In the Glashow–Weinberg–Salam model, this subsystem may appear in the form of
a condensate of W bosons. Recall that in this theory, there are three fields Aαµ , α = 1, 2,
3, and a field Bµ , from which — after symmetry breaking — the electromagnetic field

Aµ = Bµ cos θW + A3µ sin θW , (4.2.2)

 80

the massive neutral field

Zµ = Bµ sin θW − A3µ cos θW , (4.2.3)
and the charged field
1
Wµ± = √ (Aµ ∓ A2µ ) (4.2.4)
2
are formed. To be able to describe effects associated with nonvanishing lepton density,
we must append to the Lagrangian of the theory a term α l, where α is the chemical
potential corresponding to the lepton charge density. The vector field condensate that
arises at sufficiently high fermion density is of the form
W1± = C , (4.2.5)
W0± = W2± = W3± = A3i = 0 , (4.2.6)
ϕ
A30 = ± , (4.2.7)
2
where C and ϕ are determined by the equations
2 e2
* +
δL
= C2 A30 + e (neL + neR ) = 0 , (4.2.8)
δA30 sin θW
e2 Z20 e2 C2
* + !
δL 2 2
= ϕ µ − λϕ + + =0, (4.2.9)
δϕ sin2 2 θW 2 sin2 θW
e2 ϕ2 Z0
* +
δL
= + e (2neR + neL + 2 nν ) = 0 , (4.2.10)
δZ0 2 sin 2 θW
nν , neR , and neL are given by
3
1 e Z0
 
nν = α + , (4.2.11)
6 π2 sin 2 θW
1
neR = (α + e Z0 tan θW + e A0 )3 , (4.2.12)
6 π2
1
neL = (α − e Z0 cot θW + e A0 )3 . (4.2.13)
6 π2
The solution W1± = C 6= 0 can only appear at high enough lepton density, nL = nν +
neR + neL . To determine the critical value nL = ncL , one should take into account that (as
can be verified a posteriori) at nL ∼ ncL the field Z0 is of higher order in e2 than are A30 or
C. In determining ncL , we can therefore put Z0 = 0 in (4.2.8)–(4.2.13), making subsequent
analysis quite simple.
Specifically, only the trivial solution W1± = 0 exists at low densities, and we deduce
α
from (4.2.8)–(4.2.13) that A30 = A0 sin θW = − sin θW , neL = neR = 0. Starting with
e
3 α ϕ0
A0 = sin θW = , the condensate solution W1± = C 6= 0 appears. At that point
e 2
3 3
1 e ϕ0 MW

nν = ncL = = , (4.2.14)
6 π2 2 sin θW 6 π2
 81

where MW is the mass of the W boson. With a further increase in the fermion density, this
solution becomes energetically favored over the solution C = 0. As the reader can verify,
this is because the energy required to create the classical field W± is small compared with
the energy gain achieved by redistributing the lepton charge among the neutrinos and
electrons, thereby reducing the Fermi energy of the leptons.
Our final result is that both C and Z0 increase in magnitude with increasing fermion
density. When the latter is high enough, charged W± boson condensates appear. This
then leads to an asymptotic equalization of the partial densities of right- and left-handed
leptons (baryons) of various kinds in superdense matter: nνe = neR = neL , nνµ = nµR =
nµL , etc. On the other hand, according to (4.2.9), the growth of both C and Z0 will lead
to growth of the field ϕ, i.e., to the enhancement of symmetry breaking between the weak
and electromagnetic interactions.
One should note that for a particular chemical composition of cold superdense matter
4
(nB = nL , where nB and nL are the baryon and lepton densities, respectively), the
3
fermion matter proves to be neutral both with respect to the field A0 and the field Z0 .
In that case, no W condensate is formed in the superdense matter, and at high enough
density, the field ϕ will tend to vanish [29]. Interesting nonperturbative effects may
then come to the fore [174]. For the time being, we have no idea what might cause
this special regime to be realized during expansion of the universe. Strictly speaking,
4
this reservation also pertains to the more general case nB 6= nL considered above, the
3
point being that at the present time nL ∼ nB ≪ nγ . The neutrino density is not known
accurately, but in grand unified theories with nonconservation of baryon charge, it is most
reasonable to expect that at present nL ∼ nB ≪ nγ . The dominant effects, at least at
very early stages in the evolution of the universe, are then those that are related not to
the chemical potential α of cold fermionic matter, but to the temperature T ≫ α. It
is possible in principle that effects considered in this chapter may be important in the
study of certain intermediate stages in the evolution of the universe, after which there
nγ
was an abrupt increase in the specific entropy , as induced by processes considered in
nB
[97, 98, 129], for example. A combined study of both high-temperature effects and effects
related to nonvanishing lepton- and baryon-charge density, as well as an investigation of
concomitant nonperturbative effects, can be found in a number of recent papers on this
topic; for example, see [130, 175–178].
 5
Tunneling Theory and the Decay of a
Metastable Phase in a First-Order Phase
Transition

5.1 General theory of the formation of bubbles of a new phase

One important and somewhat surprising property of field theories with spontaneous
symmetry breaking is that the lifetime of the universe in an energetically unfavorable
metastable vacuum state can be exceptionally long. This phenomenon forms the basis of
the first versions of the inflationary universe scenario, according to which inflation takes
place from a supercooled metastable vacuum state (“false vacuum”) ϕ = 0 [53–55]. This
same phenomenon can lead to a partitioning of the universe into enormous, exponentially
long-lived regions in different metastable vacuum states, each corresponding to a different
local minimum of the effective potential.
For definiteness, we shall discuss the decay of the vacuum state with ϕ = 0 in the
theory with the Lagrangian
1
L(ϕ) = (∂µ ϕ)2 − V(ϕ) , (5.1.1)
2
where the effective potential V(ϕ) has a local minimum at ϕ = 0 and a global minimum
at ϕ = ϕ0 . Decay of the vacuum state with ϕ = 0 proceeds via tunneling, with the
formation of bubbles of the field ϕ. A theory of bubble production at zero temperature
was suggested in [179], and was substantially developed in [180, 181], where the Euclidean
approach to the theory of the decay of a metastable vacuum state was proposed.
We know from elementary quantum mechanics that the tunneling of a particle through
a one-dimensional potential barrier V(x) can be treated as motion with imaginary energy,
or to put it differently, as motion in imaginary time, i.e., in Euclidean space. To generalize
this approach to the case of tunneling through the barrier V (ϕ), one should consider the
wave functional Ψ(ϕ(x, t)) in place of the particle wave function ψ(x, t), and investigate
its evolution in Euclidean space. This generalization was proposed in [180, 181]. The
Euclidean approach to tunneling theory is simple and elegant, and it enables one to
progress rather far in calculating the decay probability of the false vacuum. We shall
therefore refer below (without proof) to the basic results obtained in [180, 181], and to
 83

their generalization to nonzero temperature [62]. In subsequent sections, these methods

will be applied to the study of tunneling in several specific theories.
As in conventional quantum mechanics, determination of the tunneling probability
requires first of all that we solve the classical equation of motion for the field ϕ in Euclidean
space,
d2 ϕ dV
ϕ = 2 + ∆ϕ = , (5.1.2)
dt dϕ
with boundary condition ϕ → 0 as x2 + t2 → ∞. If we then normalize V(ϕ) so that
V(0) = 0 (that is, we redefine V(ϕ) by V(ϕ) → V(ϕ) − V(0)), the tunneling probability
per unit time and per unit volume will be given by
Γ = A e−S4 (ϕ) , (5.1.3)
where S4 (ϕ) is the Euclidean action corresponding to the solution of Eq. (5.1.2)
 !2 
Z
1 dϕ 1
S4 (ϕ) = d4 x  + (∇ϕ)2 + V(ϕ) , (5.1.4)
2 dt 2

and the factor A preceding the exponential is given by

2 !−1/2
S4 det′ [− + V′′ (ϕ)]

A= . (5.1.5)
2π det[− + V′′ (0)]
d2 V
Here V′′ (ϕ) = , and the notation “det′ ” means that in calculating the functional
dϕ2
determinant of the operator [− + V′′ (ϕ)], its vanishing eigenvalues, corresponding to
the so-called zero modes of the operator, are to be omitted. This operator has four zero
modes, corresponding to the possibility of translating the solution ϕ(x) along any of the
S4 1/2
 
four axes in Euclidean space. Contributions of from each of the zero modes
2
2π
S4

result in the factor in (5.1.5).
2π
The derivation of Eqs. (5.1.3) and (5.1.5) may be found in [181], and is based on a
calculation of the imaginary part of the magnitude of the potential V(ϕ) at ϕ = 0. To
a large extent, the equations here are analogous to the corresponding expressions in the
theory of Yang–Mills instantons [182]. In essence, the solutions of Eq. (5.1.2) with the
indicated boundary conditions are scalar instantons in the theory (5.1.1). We now wish
to make several remarks before moving on to a generalization of these results to T 6= 0.
First of all, notice that in order to obtain a complete answer, it is necessary to sum over
the contributions to Γ from all possible solutions of Eq. (5.1.2). Fortunately, however, it is
usually sufficient to limit consideration to the simplest O(4)-symmetric solution ϕ(x2 +t2 ),
inasmuch as these are usually the very solutions that minimize the action S4 . In that event,
Eq. (5.1.2) takes on an even simpler form,
d2 ϕ 3 dϕ
+ = V′ (ϕ) , (5.1.6)
dr 2 r dr
 84

√ dϕ
where r = x2 + t2 , with boundary conditions ϕ → 0 as r → ∞ and = 0 at r = 0.
dr
2 2
The high degree of symmetry inherent in the solution ϕ(x + t ) helps us to obtain
a graphic description of the structure and evolution of a bubble of the field ϕ after it is
created. To do so, we analytically continue the solution to conventional time, t → i t,
or in other words ϕ(x2 + t2 ) → ϕ(x2 − t2 ). Since the solution ϕ(x2 − t2 ) depends only
on the invariant quantity x2 − t2 , the corresponding bubble will look the same in any
reference frame, and the expansion speed of a region filled with the field ϕ (the speed
of the bubble “walls” will asymptotically approach the speed of light. The creation and
growth of bubbles is an interesting mathematical problem [180], but one that we shall not
discuss here, as our main objective is to study those situations in which the probability
of bubble creation is negligible.
Unfortunately, Eq. (5.1.6) can seldom be solved analytically, so that both the solution
and the associated value of the Euclidean action S4 (ϕ) must often be computed numer-
ically. In this sort of situation, determinants can only be calculated in certain special
cases. It turns out, however, that in most practical problems just a rough estimate of the
pre-exponential factor A will suffice. We can come up with such an estimate by noting
that the factor A has dimensionality m4 , and its q value is determined by three different
quantities with dimensionality m, namely ϕ(0), V′′ (ϕ), and r −1 , where r is the typical
size of a bubble. In the theories that interest us most, all of these quantities lie within an
order of magnitude of one another, so for a rough estimate one may assume that
det′ [− + V′′ (ϕ)]
= O(r −4 , ϕ4 (0), (V′′ )2 ) , (5.1.7)
det[− + V′′ (0)]
where we denote by r and V′′ (ϕ) typical mean values of these parameters for the solution
ϕ(r) of Eq. (5.1.6). Next, let us proceed to the case in which T 6= 0 [62]. In order to
generalize the preceding results to this instance, it suffices to recall that the quantum
statistics of bosons (fermions) at T 6= 0 are formally equivalent to quantum field theory
in a Euclidean space with a periodicity (antiperiodicity) of 1/T in the “time” β (see [166],
for example). When one considers processes at fixed temperature, the quantity V(ϕ, T)
plays the role of the potential energy. The imaginary part of this function in an unstable
vacuum can be calculated in just the same way as is done in [181] for the case T = 0.
The only essential difference is that instead of finding an O(4)-symmetric solution of Eq.
(5.1.2), one must find an O(3)-symmetric (in the spatial coordinates) solution periodic in
the “time” β, with period 1/T. As T → 0, the solution of Eq. (5.1.2) that minimizes the
action S4 (ϕ) consists of an O(4)-symmetric bubble with some typical radius r(0) (Fig.
5.1a). As T → r −1 (0), the solution becomes a series of such bubbles separated from one
another by a distance 1/T in the “time” direction (Fig. 5.1b). When T ∼ r −1 (0), the
bubbles start to overlap (Fig. 5.1c). Finally, when T ≫ r −1 (0) (which is just the case
that is of most importance and interest to us), the solution becomes a cylinder whose
spatial cross section is an O(3)-symmetric bubble of some new radius r(T) (Fig. 5.1d).
When we calculate S4 (ϕ) in the latter case, the integration over β reduces simply to a
1
multiplication by 1/T — that is, S4 (ϕ) = S3 (ϕ), where S3 (ϕ) is the three-dimensional
T
 85

∫
β ∫β β β
4
T
1 2
T 3
T T
2
1 T
r(0) T r(T)
1
T
0 r 0 r 0 r 0 r

1
1 T
T 2
T
1 2 3
T T T
4
T
3
T 5
T

a b c d

Figure 5.1: The form taken by the solution to Eq. (5.1.2) at various temperatures: a)
T = 0; b) T ≪ r −1 (0); c) T ∼ r −1 (0); d) T ≫ r −1 (0). The shaded regions contain the
classical field ϕ 6= 0. For simplicity, we have drawn bubbles for those cases in which the
thickness of their walls is much less than their radii.

action corresponding to the O(3)-symmetric bubble,

1
Z  
3
S3 (ϕ) = d x (∇ϕ)2 + V(ϕ, T) . (5.1.8)
2
To calculate S3 (ϕ), we must solve the equation

d2 ϕ 2 dϕ dV(ϕ, T)
2
+ = = V′ (ϕ, T) (5.1.9)
dr r dr dϕ
dϕ
with boundary conditions ϕ → 0 as r → ∞ and = 0 as r → 0. In the high-temperature
dr
limit (T ≫ r −1 (0)), the complete expression for the tunneling probability per unit time
and per unit volume is obtained in a manner completely analogous to that employed in
[181] to derive Eqs. (5.1.4) and (5.1.5); the result is
!
S3 (ϕ, T)
4
Γ(T) ∼ T exp − . (5.1.10)
T
 86

ϕ1 ϕ0
0
ϕ
−ε
Figure 5.2: Effective potential V(ϕ) in the case of slight supercooling of the phase ϕ = 0
(i.e., the quantity ε = V(0, T) − V(ϕ0 , T) is small).

It can be seen from Eq. (5.1.10) that the main problem to be solved in determining
the probability of bubble creation is to find S3 (ϕ, T) (or S4 at T = 0). Furthermore, if we
are to obtain reasonable estimates of the determinants, and wish to be able to study the
expansion of the bubbles that are formed, we must know the form taken by the function
ϕ(r) and the typical size of a bubble. As we pointed out earlier, the corresponding results
are obtained, as a rule, by computer solution of the equations, which seriously complicates
the investigation of phase transitions in realistic theories. It is therefore of particular
interest to study those instances in which the problem can be solved analytically, and we
treat one such case in the next section. From here on, we shall consider not only the
case T ≫ r −1 (0), but the case T = 0 as well, as the latter gives us information on the
probability of bubble creation in the limit of a strongly supercooled metastable phase,
where T ≪ r −1 (0).

5.2 The thin-wall approximation

In tunneling theory, there are two limiting cases in which the problem simplifies consid-
erably. One of these is associated with the situation where V(ϕ) at a minimum with
ϕ = ϕ0 (τ ) 6= 0 is much larger in absolute value than the height of the potential barrier in
V(ϕ) between ϕ = 0 and ϕ = ϕ0 ; that case will be taken up in the next chapter. Here we
examine the other limit, in which |V(ϕ0 )| = ε is much lower than the barrier height (see
Fig. 5.2).
It is readily seen that as decreases, the volume energy involved in bubble creation
(∼ ε r 3) becomes large compared to the surface energy (∼ r 2 ) for large r only if the bubble
is big enough. When the size of a bubble greatly exceeds the bubble wall thickness (the
dϕ
wall being that region where derivatives are large), one can neglect the second term
dr
in (5.1.6) and (5.1.9) compared with the first. In other words, these equations effectively
 87

ϕ0

0 r(T) r
Figure 5.3: Typical form of the solution of Eqs. (5.1.6) and (5.1.9) when ε → 0.

reduce to one that describes tunneling in one-dimensional space-time:

d2 ϕ
= V′ (ϕ, T) . (5.2.1)
dr 2
In the limit as ε → 0, the solution of this equation takes the form
ϕ0 dϕ
Z
r= q , (5.2.2)
ϕ 2 V(ϕ)

where the functional form of ϕ(r) has been sketched in Fig. 5.2.
Let us first consider tunneling in quantum field theory (T = 0). In an O(4)-symmetric
bubble (5.2.2), the action S4 is given by
 !2 
∞ 1 dϕ
Z
S4 = 2 π 2 r 3 dr  + V
0 2 dr
ε
= − π 2 r 4 + 2 π 2 r 3 S1 , (5.2.3)
2
where S1 is the surface energy of the bubble wall (surface tension), and is equal to the
action in the corresponding one-dimensional problem (5.2.1):
 !2 
Z ∞ 1 dϕ
S1 = dr  + V
0 2 dr
Z ϕ0 q
= dϕ 2 V(ϕ) , (5.2.4)
0

the integral in (5.2.4) being calculated in the limit as ε → 0. We find the bubble radius
r(0) by minimizing (5.2.3):
3 S1
r(0) = , (5.2.5)
ε
 88

whereupon
27 π 2 S41
S4 = . (5.2.6)
2 ε3
Notice that to order of magnitude, the bubble wall thickness is simply (V′′ (0))−1/2 . Taking
(5.2.5) into account, therefore, the condition for the present approximation (the so-called
thin-wall approximation) to be valid is

3 S1
≫ (V′′ (0))−1/2 . (5.2.7)
ε
The foregoing results were derived by Coleman [180]. We can now readily generalize these
results to the high-temperature case, T ≫ r −1 (0). To do so, we merely point out that
 !2 
Z ∞ 1 dϕ
S3 = 4 π r 2 dr  + V(ϕ, T)
0 2 dr
4π 3
= − r ε + 4 π r 2 S1 (T) , (5.2.8)
3
so that
2 S1
r(T) = (5.2.9)
ε
and
16 π S31
S3 = . (5.2.10)
3 ε2
The expression thus obtained for the probability of bubble formation,

16 π S3
!
Γ ∼ exp − 2 1 , (5.2.11)
3ε T

is consistent with the well-known expression found in textbooks [61]. The only difference
(but an important one) is that we have the closed expression (5.2.4) for the surface tension
S1 , where instead of V(ϕ) one should use V(ϕ, T). In many cases of interest, the function
V(ϕ, T) plotted in Fig. 5.2 can be approximated by the expression

M 2 δ 3 λ 4
V(ϕ) = ϕ − ϕ + ϕ . (5.2.12)
2 3 4
Let us investigate bubble formation in this theory in more detail, since for the potential
(5.2.12) one can evaluate the integral in (5.2.4) exactly, and it thereby becomes possible
to obtain analytic expressions for S1 , S3 , S4 and r(T).
In fact, it can readily be demonstrated that for values of the parameters M, δ, and
λ such that the minima at ϕ = 0 and ϕ = ϕ0 are of equal depth (ε → 0), Eq. (5.2.12)
becomes
λ
V(ϕ) = ϕ2 (ϕ − ϕ0 )2 , (5.2.13)
4
 89

and in that event ϕ0 is

2δ
ϕ0 = , (5.2.14)
λ
while M, δ, and λ are related by
2 δ 2 = 9 M2 λ . (5.2.15)
From (5.2.8) and (5.2.13)–(5.2.15), it follows that
s
λ ϕ30
S1 = = 23/2 3−4 δ 3 λ−5/2 , (5.2.16)
2 6
whereupon for T = 0 one obtains
π 2 25 δ 12 23/2 δ 3
S4 = , r(0) = , (5.2.17)
313 λ10 ε3 33 λ5/2 ε
while for T ≫ r −1 (0),
217/2 π δ 9 25/2 δ 3
S3 = , r(T) = . (5.2.18)
313 λ15/2 ε2 34 λ5/2 ε
We now turn to the specific case of phase transitions in gauge theories at high tem-
perature. Here a typical expression for V(ϕ, T) is
β (T2 − T2c1 ) 2 α λ
V(ϕ, T) = ϕ − T ϕ3 + ϕ4 , (5.2.19)
2 3 4
where Tc1 is the temperature above which the symmetric phase ϕ = 0 is metastable, and
β and α are numerical coefficients (compare (3.1.21), (3.1.22)). The temperature Tc at
which the values of V(ϕ, T) for the phases with ϕ = 0 and ϕ = ϕ0 (T) are equal is given
by !−1
2 2 2 α2
Tc = Tc1 1 − . (5.2.20)
9βλ
One can readily determine the quantity e as a function of the departure of the temperature
T from its equilibrium value:
4 Tc T2c1 α2 β
ε= ∆T , (5.2.21)
9 λ2
where ∆T = Tc − T. It is then straightforward, using Eqs. (5.2.14)–(5.2.20), to derive
expressions for the quantities of interest, namely S3 and r(T). These may be written out
Tc − T
for the most frequently encountered situation, with x = ≪ 1:
Tc
S3 29/2 π α5 1
S4 = = 9 2 7/2 2 , (5.2.22)
T 3 β λ x
s
2 α 1
r = . (5.2.23)
λ 9 β Tc x
 90

Thus, the thin-wall approximation makes it possible to progress rather far towards an
understanding of the formation of bubbles of a new phase. Unfortunately, however, this
method can only be applied to relatively slow phase transitions, or to be more precise,
those for which
S3
S4 = >
∼ 10 α λ−3/2 . (5.2.24)
T
This restriction is not satisfied in many cases of interest, forcing us to seek ways of
proceeding beyond the scope of the thin-wall approximation.

5.3 Beyond the thin-wall approximation

We have already remarked that there is one more instance in which the theory of bubble
creation may be considerably simplified. Specifically, if the minimum of V(ϕ) at the point
ϕ0 is deep enough, the maximum value of the field ϕ(r) corresponding to the solution of
Eqs. (5.1.6) and (5.1.9) becomes of order ϕ1 , where V(ϕ1 ) = V(0), ϕ1 ≪ ϕ0 . In solving
(5.1.6) and (5.1.9), one can then neglect the details of the behavior of V(ϕ) for ϕ ≫ ϕ1 ,
and when ϕ < ∼ ϕ1 , it is often possible to approximate the potential V(ϕ) with one of two
basic types of functions:
M2 2 λ 4
V1 (ϕ) = ϕ − ϕ , (5.3.1)
2 4
2
M δ
V2 (ϕ) = ϕ2 − ϕ3 . (5.3.2)
2 3
At zero temperature and with M = 0, Eq. (5.1.6) for the theory (5.3.1) can be solved
exactly [182], s
8 ρ
ϕ= , (5.3.3)
λ r + ρ2
2

where ρ is an arbitrary parameter with dimensionality of length (the arbitrariness in the

choice of ρ is a consequence of the absence of any mass parameter in the theory (5.3.1)
for M = 0). For all ρ, the action corresponding to the solutions of (5.3.3) is

8 π2
S4 = . (5.3.4)
3λ
To find the total probability of bubble formation, one must integrate (with a certain
weight) the contributions from solutions (instantons) for all values of ρ, as in the theory
of Yang–Mills instantons [183].
At T = 0 and arbitrary M 6= 0, Eq. (5.1.6) in the theory (5.3.1) has no exact
instanton solutions of the type we have studied [184], for the same reason that there are
no instantons in the theory of massive Yang–Mills fields. On the other hand, for ρ ≪ M−1 ,
the solution of (5.3.3) is essentially insensitive to the presence of a mass M in the theory
(5.3.1). Therefore, for T = 0 and M 6= 0, the theory (5.3.1) admits of “almost exact
 91

φ
6 5.78

5
B

4
3.08

3 2.78

D
2

C
1
A

0 1 2 3 4 R

Figure 5.4: The form of bubbles ϕ(r) in the theories (5.3.1) and (5.3.2) at T = 0 and
T ≫ r −1 (0). The behavior of ϕ as a function of r has been plotted in this figure in terms
of the dimensionless variables R = r M and Φ = ϕ/ϕ1 , where ϕ1 is defined by V(ϕ1 , T) =
V(0, T). Curves A and B are O(4)-symmetric bubbles in the theories (5.3.1) and (5.3.2)
respectively; curves C and D are O(3)-symmetric bubbles in those same theories.

solutions” of Eq. (5.1.6) that are identical to (5.3.3) when ρ ≪ M−1 . This means that
an entire class of trajectories (5.3.3) exists in Euclidean space that describes formation of
a bubble of the field ϕ 6= 0. To high accuracy, the action corresponding to each of these
trajectories in the theory (5.3.1) with M 6= 0 is the same as (5.3.4) when ρ ≪ M−1 , and
it tends to the minimum (5.3.4) as ρ → 0. As a result, tunneling does exist at T = 0 in
the theory (5.3.1), and to describe it, one must integrate over ρ the contributions to Γ
from all “solutions” (5.3.3) with ρ ≪ M−1 , as is done in the theory of instantons when
the Yang–Mills field acquires mass [183]. In this somewhat inexact sense, we will speak
of solutions of Eq. (5.1.6) in the theory (5.3.1) with M 6= 0 and with the action (5.3.4) (a
similar situation is investigated in [185]).
In all other cases under consideration (at high temperature in the theory (5.3.1), and
at both high and low temperature in the theory (5.3.2)), exact solutions do exist. The
form taken by the solutions is depicted in Fig. 5.3. At T = 0, the action S4 corresponding
to the solution ϕ(r) in the theory (5.3.2) is
M2
S4 (ϕ) ≈ 205 . (5.3.5)
δ2
 92

S3
In the high-temperature limit, the action S4 (ϕ) = for the solutions corresponding to
T
the theories (5.3.1) and (5.3.2) is

19 M
S4 (ϕ) ≈ (5.3.6)
λT
and
M3
S4 (ϕ) ≈ 44 (5.3.7)
δ2 T
respectively.
Note that the results obtained above do not just refer to the limiting cases T = 0 and
T ≫ M. An analysis of this problem shows that Eqs. (5.3.4) and (5.3.5) continue to hold
down to temperatures T ≤ 0.7 M (T ≤ 0.2 M), and at higher temperatures one can make
use of the results (5.3.6) and (5.3.7) [62].
To conclude this chapter, let us consider briefly the most typical case, in which the
potentials V1 and V2 are of the form

β (T2 − T2c1 ) 2 λ 4
V1 (ϕ, T) = ϕ − ϕ , (5.3.8)
2 4
2 β (T2 − T2c1 ) 2 α
V (ϕ, T) = ϕ − T ϕ3 . (5.3.9)
2 3
From the previous results, it follows that at high enough temperature in the theory (5.3.8),
q
19 β (T2 − T2c1 )
S4 = , (5.3.10)
λT
while in the theory (5.3.9),

44 [β (T2 − T2c1 )]3/2

S4 = . (5.3.11)
α2 T3
In many realistic situations, the effective potential near a phase transition point is
well-approximated by one of the types considered in Sections 5.2 and 5.3. The results
obtained above may therefore often be directly applied to studies of the kinetics of the
first-order phase transitions in realistic theories. We shall use these results to analyze a
number of specific effects in Chapters 6 and 7.
At this point, we would like to add two remarks in connection with the foregoing
results. We see from Eqs. (5.3.4)–(5.3.7) that for certain values of the parameters that
enter into these equations, the probability that a metastable phase will decay can be ex-
ceedingly low. For example, when λ ∼ 10−2 , tunneling in the theory (5.3.1) is suppressed
by a factor
8 π2
!
P ∼ exp − ∼ exp(−103 ) . (5.3.12)
3λ
 93

This explains why in realistic theories, metastable vacuum states can turn out to be
almost indistinguishable from the stable one. In particular, one has essentially no reason
to think that the vacuum state in which we now reside is the one corresponding to the
absolute minimum of energy. One might try, in principle, to carry out an experiment to
test the stability of our vacuum by attempting to create a nucleus of a new phase (e.g.,
via heavy ion collisions), but both the technological feasibility and, understandably, the
advisability of such an experiment are highly dubious.1
The second remark bears on the range of applicability of the foregoing results. These
were obtained by neglecting effects associated with the expansion of the universe, an
approximation that is perfectly adequate if the curvature V′′ (ϕ) of the effective potential
is much greater than the curvature tensor Rµναβ . But in the inflationary universe scenario,
V′′ (ϕ) ≪ R = 12 H2 during inflation. Tunneling during inflation must therefore be studied
as a separate issue. We shall return to this question in Chapter 7.

1
One might argue, of course, that such an experiment would be quite enlightening, regardless of the
results. If the experiment were to confirm the stabilitty of our vacuum state, it would make us all very
proud. On the other hand, if a bubble of a more energetically advantageous vacuum state were produced,
the observable part of the universe would gradually be transformed into a better vacuum state, and no
observes would remain to be dissatisfied with the experimental results.
 6
Phase Transitions in a Hot Universe

6.1 Phase transitions with symmetry breaking between the weak, strong,
and electromagnetic interactions
We have already pointed out in Chapter 1 that according to the standard hot universe
theory, the expansion of the universe started from a state of enormously high density, at
a temperature T much higher than the critical temperature of a phase transition with
symmetry restoration between the strong and electroweak interactions in grand unified
theories. Therefore, the symmetry between these interactions should have been restored
in the very early stages of the evolution of the universe.
As the temperature decreases to T ∼ Tc1 ∼ 1014 –1015 GeV (see Eq. (3.2.6)), a
phase transition (or several) takes place, generating a classical scalar field Φ ∼ 1015 GeV,
which breaks the symmetry between the strong and electroweak interactions. When the
temperature drops to Tc2 ∼ 200 GeV (see Eq. (??), the symmetry between the weak and
electromagnetic interactions breaks. Finally, at T ∼ 102 MeV, there should be a phase
transition (or two separate transitions) which breaks the chiral invariance of the theory
of strong interactions and leads to the coalescence of quarks into hadrons (confinement).
Here we must voice some reservations. The Glashow–Weinberg–Salam theory of elec-
troweak interactions has withstood experimental tests quite well, but the situation with
grand unified theories is not nearly so satisfactory. Prior to the 1980’s, there seemed to be
little doubt of the existence of grand unification at energies E ∼ 1015 GeV, with the most
likely candidate for the role of a unified theory being minimal SU(5). Subsequently, uni-
fied theories became more and more complicated, starting with N = 1 supergravity, then
the Kaluza–Klein theory, and finally superstring theory. As the theories have changed,
so has our picture of the evolution of the universe at high temperatures. But all versions
of this picture have at least one thing in common: without an inflationary stage, they
all lead to consequences in direct conflict with existing cosmological data. In the present
section, in order to expose the sources of these problems and point out some possibilities
for overcoming them, we will study the kinetics of phase transitions in minimal SU(5)
theory.
In that theory, the potential in the field Φ responsible for symmetry breaking between
 95

the strong and electroweak interactions takes the form (see Section 3.2)

µ2 a b
V(Φ) = − Tr Φ2 + (Tr Φ2 )2 + Tr Φ4 . (6.1.1)
2 4 2
At T ≫ µ, the main modification to V(Φ) consists of a change of sign of the effective
parameter µ2 ,
µ2 (T) = µ2 − β T2 , (6.1.2)
see (3.2.3). This leads to the restoration of symmetry at high temperatures. According
to (3.2.3), however, at T < ∼ µ the modification of the effective potential does not reduce
2
to a change in µ ; the effective potential V(Φ, T) can acquire additional local minima
that correspond not just to SU(3) × SU(2) × U(1) symmetry breaking (see Chapter 1),
but also to symmetry breaking described by the groups SU(4) × U(1), SU(3) × (SU(1))2 ,
or (SU(2))2 × (SU(1))2 [167]. This, plus the fact that phase transitions in grand unified
theories are first-order transitions, greatly complicates investigation of the kinetics of the
transition from the SU(5) phase to the SU(3) × SU(2) × U(1) phase. Here we present the
main results from this investigation [187].
First of all, recall that according to [167], the effective potential V(ϕ, T) of the minimal
SU(5) theory takes the form

N π T4 µ2 (T) 2
V(ϕ, T) = − − ϕ − αi T ϕ3 + γi ϕ (6.1.3)
90 2
for each of the four types of symmetry breaking mentioned above, where ϕ2 = Tr Φ2 , and
αi and γi, i = 1, 2, 3, 4 are certain constants calculated in [167]. This effective potential
is the same as the potential (5.2.12), so that all of the results we obtained in the thin-wall
approximation concerning tunneling from a state ϕ 6= 0, with formation of bubbles of a
field ϕ 6= 0, also apply to the theory (6.1.3). On the other hand, in those cases where
the thin-wall approximation does not work, the field ϕ within a bubble is small, the last
term in (6.1.3) can be discarded, and the potential is the same as (5.3.2), for which we
also studied tunneling in Chapter 5.
Our plan of attack is thus as follows. We must understand how the quantity V(ϕ, T)
in (6.1.3) depends on time in an expanding universe, calculate the rate of production of
each of the four types of bubbles enumerated above, determine the moment at which the
bubbles thus formed occupy the whole universe, explore what happens to bubbles formed
in earlier stages, and find the typical volume occupied by regions that are filled with the
various phases at the end of the whole process.
Since we have already developed the theory of bubble formation, the solution of the
foregoing problem should not be particularly difficult. Nevertheless, it does turn out to be
a fairly tedious task, since the numerical calculations must be rerun for every new choice
of parameters a and b in (6.1.1). Below we present and discuss the main results that we
have obtained for the most natural case, a ∼ b ∼ 0.1 [187].
For this case, the phase transition takes place from a supercooled state in which the
temperature of the universe approaches Tc1 ; starting with this temperature, the symmetric
 96

phase ϕ becomes absolutely unstable. The jump in the field ϕ at the phase transition
point is then large (of order ϕ0 ). In that sense, the phase transition is a “strong” first-order
transition.
The phase transition proceeds with the simultaneous production of all four types of
phases listed above, the overwhelming majority of the bubbles containing the SU(4)×U(1)
phase, and not the energetically more favorable SU(3)×SU(2)×U(1) phase, which initially
occupies only a few percent of the whole volume. SU(3)×SU(2)×U(1) bubbles eventually
start expanding within the SU(4) × U(1) phase, “devouring” both it and the bubbles of
the other two phases. At such time as the SU(3) × SU(2) × U(1) bubbles coalesce, they
have a typical size of
r ∼ T−1
c1 . (6.1.4)
Prior to the formation of a homogeneous SU(3) × SU(2) × U(1) phase, the kinetics of
processes during the intermediate phase is very complex, depending on the values of a,
b, and g 2 . The duration of this intermediate stage, as well as that of the stage preceding
the end of the phase transition, can only be significant in theories with certain specific
relations between the coupling constants.
Despite the large jump in the field ϕ at the phase transition point, the amount of energy
liberated in the phase transition process is relatively minor, as a rule, so that given the
most reasonable values of the coupling constants, a symmetry-breaking transition from a
supercooled SU(5)-symmetric phase will not result in a discontinuous rise in temperature,
nor will it produce a marked increase in the total entropy of the expanding universe.
As the temperature drops further to Tc2 ∼ 102 GeV, the phase transition SU(3) ×
SU(2)×U(1) → SU(3)×U(1) takes place, and with it the symmetry between the weak and
electromagnetic interactions is broken. At the time of this transition, the temperature is
many orders of magnitude lower than the mass of the superheavy bosons with MX ∼ 1014
GeV that appear after the first phase transition. Lighter particles in this theory are
described by the Glashow–Weinberg–Salam theory, so the phase transition at Tc2 ∼ 102
GeV proceeds exactly as in the latter; see Chapter 3.
Generally speaking, the foregoing pattern of phase transitions is only relevant to the
simplest grand unified theories with the most natural relation between coupling constants.
In more complicated theories, phase transitions may occur with many more steps; for
example, see [42, 167]. A somewhat unusual picture also emerges for certain special
relations among the parameters of a theory, for which imply the effective potential of
scalar fields contains a local minimum or a relatively flat region at small ϕ.
By way of example, let us consider the Glashow–Weinberg–Salam model with
1 d4 V  11 e4 2 cos4 θW + 1

λ(ϕ0 ) = < · ≈ 3 · 10−3 , (6.1.5)
6 dϕ4 ϕ=ϕ0 16 π 2 sin2 2 θW
d2 V  e4 ϕ20 2 cos4 θW + 1

m2ϕ (ϕ0 ) = < ·
dϕ2 ϕ=ϕ0 16 π 2 sin2 2 θW
≈ (10 GeV)2 , (6.1.6)

where sin2 θW ≈ 0.23, ϕ0 ≈ 250 GeV. For these values of λ(ϕ0 ) and m2ϕ , the effective
 97

potential V(ϕ) has a local minimum at ϕ = 0 even at zero temperature [139–141]; see
Section 2.2.
In that case, symmetry was restored in the early universe as usual, with ϕ = 0. As
the universe cooled, a minimum of V(ϕ) then appeared at ϕ ∼ ϕ0 , becoming deeper than
the one at ϕ = 0 shortly thereafter. Nevertheless, the universe remained in the state
ϕ = 0 until such time as bubbles of a new phase with ϕ 6= 0 formed and filled the entire
universe. The formation of bubbles of a new phase in the Glashow–Weinberg–Salam
theory was studied in [141, 142]. It turns out that if mϕ is even one percent less than
the limiting value mϕ ∼ 10 GeV (6.1.6), the probability of bubble formation with ϕ 6= 0
becomes exceedingly small.
The reason for this is not far to seek if we hark back to the results of the previous
chapter. Consider the limiting case
e4 ϕ20 2 cos4 θW + 1
m2ϕ = · . (6.1.7)
16 π 2 sin2 2 θW
The curvature of V(ϕ) at ϕ = 0, T = 0 then tends to zero (the Coleman–Weinberg model
[137]; see Section 2.2). At T 6= 0, the mass of the scalar field in the vicinity of ϕ = 0 is,
according to (??),
eT q
mϕ ∼ 1 + 2 cos2 θW (6.1.8)
sin 2 θW
(recall that in the present case λ ∼ e4 ≪ e2 ). In this model, at small ϕ, the potential
V(ϕ) is approximately
3 e4 ϕ4 2 cos4 θW + 1 m2 ϕ2
!
ϕ
V(ϕ) = V(0) + ln + ϕ . (6.1.9)
32 π 2 sin2 2 θW ϕ0 2
ϕ
Now ln is a fairly slowly varying function of ϕ, so to determine the probability P of
ϕ0
tunneling out of the local minimum at ϕ = 0, we can make use of Eq. (5.3.6) [144]:
 
!
19 mϕ (T) 19 sin 2 θW
 
 
P ∼ exp − −
∼ exp  
λT 3 e3 q ϕ 
1 + cos4 θW ln
 
8π 2 ϕ0
 
 15000 
∼ exp 
− ϕ  .
 (6.1.10)
ln
ϕ0
The typical value of the field ϕ appearing in (6.1.10) corresponds to a local maximum of
V(ϕ) in (6.1.9) located at ϕ ∼ 10 T; that is,
 
 15000 
P ∼ exp −  . (6.1.11)
 
T 
ln

ϕ0
 98

Hence, we find that in the theory under consideration, the phase transition in which
bubbles of the field ϕ are formed can only take place if the temperature of the universe
is exponentially low. A similar phenomenon in the Coleman–Weinberg SU(5) theory
lays at the basis of the new inflationary universe scenario (see Chapter 8). But in the
Glashow–Weinberg–Salam theory with

d2 V 

=0,
dϕ2 ϕ=0

supercooling is actually not so strong as might be construed from

(6.1.11): the phase transition occurs at T ∼ 102 MeV on account of effects associated
with strong interactions [144]. When it takes place, the specific entropy of the universe
nγ
should rise approximately 105 –106 -fold [144], due to liberation of the energy stored in
nB
the metastable vacuum ϕ = 0. Even if the effective potential V(ϕ) has only a very shal-
low minimum at ϕ = 0, the increase in the specific entropy of the universe may become
unacceptably large [143, 144]. Furthermore, the lifetime of the universe in a metastable
2 2
vacuum state with V′′ (0) >∼ (10 MeV) will be greater than the age of the observable
part of the universe, t ∼ 1010 yr [141, 142]. Bubbles formed as a result of such a phase
transition would make the universe strongly anisotropic and inhomogeneous. The uni-
verse would be homogeneous only inside each of the bubbles, which would be devoid of
matter of any kind. This leads to the strong constraint (2.2.14) on the mass of the Higgs
boson in the Glashow–Weinberg–Salam theory without superheavy fermions1 :

mϕ >
∼ 10 GeV . (6.1.12)

As we showed in Chapter 2, the absolute minimum of V(ϕ) in a theory with superheavy

µ
fermions may turn out not to be at ϕ = ϕ0 = √ , but at ϕ ≫ ϕ0 , which constrains the
λ
allowable fermion masses in the theory [146–151]. When cosmological effects are taken
into account, the corresponding bounds are softened somewhat, since the universe will
not always succeed in going from a state ϕ = ϕ0 to an energetically more favorable
one [188]. The complete set of bounds on the masses of fermions and the Higgs boson,
including the cosmological constraints, is shown in Fig. 2.2 (Chapter 2, page 59). Notice,
however, that in studying tunneling, the authors of [141–151, 188] did not discuss the
possibility of tunneling induced by collisions of cosmic rays with matter. If such processes
could substantially increase the probability of decay of a metastable vacuum [189] (see,
however, [360]), then the region above the curve AD in Fig. 2.2 would turn out to be
forbidden, and the most stringent constraint on mϕ would be set by Eq. (2.2.9). This
problem requires more detailed investigation.
1
To avoid misunderstanding, we should emphasize that these bounds refer only to the simplest version
of the Glashow–Weinberg–Salam theory, with a single type of scalar field ϕ
 99

6.2 Domain walls, strings, and monopoles

In the preceding section, we pointed out that a phase transition with SU(5) symmetry
breaking takes place with the formation of bubbles containing several different phases, and
only subsequently does all space fill with matter in a single energetically most favorable
phase. For this to happen, at least two conditions must be satisfied: only one energetically
most favorable phase may exist, and the typical size r of the bubbles must not exceed
t, where t is the time at which the entire universe should have made the transition to a
single phase. In the hot universe theory (in contrast to the inflationary universe theory),
bubbles typically do not grow to be very large — r ∼ m−1 or r ∼ T−1 (see Fig. 5.3, page
91) — so the second requirement is usually met. But there are a great many theories
in which the effective potential has several minima of the same (or almost the same)
µ
depth. The simplest example is the theory (1.1.5), which has minima at ϕ = √ and
λ
µ
ϕ = − √ of equal depth. When a phase transition occurs at some time t = tc during
λ
the expansion of the universe, symmetry breaking takes place independently in different
causally disconnected regions of size O(tc ). As a result, the universe is partitioned into
µ µ
approximately equal numbers of regions filled with the fields ϕ = √ and ϕ = − √ .
λ λ
These regions are separated from one another by domain walls of thickness O(µ−1 ), with
µ µ
the field changing from ϕ = √ to ϕ = − √ from one side of the wall to the other.
λ λ
Actually, as a rule, the regions in which symmetry breaking takes place independently
initially have sizes of order T−1c ; that is, they have dimensions much smaller than the
M P
horizon t ∼ 10−2 2 at the time when the phase transition starts. One example of this
Tc
is the formation of regions with different phases at the time of the phase transition in the
SU(5) model; see (6.1.4).
Regions that are filled with different phases at the same energy density also tend to
“eat” each other, as the presence of domain walls is energetically unfavorable. But this
mutual consumption proceeds independently in regions separated by distances of order t,
where t is the age of the universe. As we have already noted in Section 1.5, at time t ∼ 105
yr, the presently observable part of the universe consisted of approximately 106 causally
disconnected regions, or, in other words, of 106 domains separated by superheavy domain
walls. Since in the last ∼ 105 years the observable part of the universe has been transparent
to photons, the existence of such domains would lead to considerable anisotropy in the
primordial background radiation. The astronomical observations, however, indicate that
∆T
the background radiation is isotropic to within ∼ 3 · 10−5. This is the essence
T
of the domain wall problem in the hot universe theory [41]. These results would force
us to renounce theories with discrete symmetry breaking, such as the theory (1.1.5),
theories with spontaneously broken CP invariance, the minimal SU(5) theory, in which
the potential V(Φ) (6.1.1) is invariant under reflection Φ → −Φ, etc. Most theories of
the axion field θ encounter similar difficulties: the potential V(θ) in many versions of the
axion theory has several minima of the same depth [49]. In some theories, this difficulty
 100

Figure 6.1: Distribution of the field χ = ϕ(x) eiθ(x) in isotopic space over a path surround-
ing a string ϕ(x) = 0.

can be overcome (for example, by adding a term c Tr Φ3 to V(Φ) (6.1.1)), but usually the
problems are insurmountable without changing the theory fundamentally (or reverting to
the inflationary universe scenario).
Besides domain walls, phase transitions can give rise to other nontrivial entities as
well. Consider, for example, a model of a complex scalar field χ with the Lagrangian

L = ∂µ χ∗ ∂µ χ + m2 χ∗ χ − λ (χ∗ χ)2 . (6.2.1)

This is the Higgs model of (1.1.15) prior to the inclusion of the vector fields Aµ . In order
to study symmetry breaking in this theory, it is convenient to change variables:
!
1 i ζ(x)
χ(x) → √ ϕ(x) exp . (6.2.2)
2 ϕ0
µ
The effective potential V(χ, χ∗ ) has a minimum at ϕ(x) = ϕ0 = √ irrespective of
λ
the value of the constant part of the phase ζ0 . V(χ, χ∗ ) is thus shaped like the bot-
tom of a basin, with a maximum in the middle (at χ(x) = 0), and rather than being
characterized simply
! by the scalar ϕ0 , symmetry breaking is characterized by the vector
i ζ(x)
ϕ(x) exp in the (χ, χ∗ ) isotopic space.
ϕ0
The existence of fields with different phases ζ(x) in different regions of space is energet-
ically unfavorable. But just as in the case of domain! walls, the value of the phase — that
i ζ(x)
is, the direction of the vector ϕ(x) exp — cannot be correlated over distances
ϕ0
greater than the size of the horizon, ∼ t. Moreover, immediately after the phase transi-
tion, the direction of this vector at different points x cannot be correlated over distances
much greater than O(T−1 c ).
Let us consider some two-dimensional surface in our three-dimensional space and study
the possible configurations of the field ϕ there. Among these configurations, there is one
 101
!
i ζ(x)
such that upon traversing some closed contour in x-space, the vector ϕ(x) exp
ϕ0
ζ(x)
executes a complete rotation in (χ, χ∗ ) isotopic space (i.e., the function changes by
ϕ0
2π); see Fig. 6.2. The appearance of such an initial field distribution for ϕ as a result
of a phase transition is in no way forbidden. Now let us gradually constrict this contour,
remaining all the while in a region with ϕ(x) 6= 0. Since the field χ(x) is continuous and
differentiable, the vector χ(x) should also execute a complete rotation in traveling along
the shrinking contour. If we could shrink the contour to a point at which ϕ(x) 6= 0 in
this manner, the field χ(x) would no longer be differentiable there; that is, the equations
of motion would not hold at that point. This implies that somewhere within the original
contour there must be a point at which ϕ(x) = 0. Let us suppose for the sake of simplicity
that there is just one. Now change the section of space under consideration, appropriately
moving our contour in space so that as before it does not pass through any region with
ϕ(x) = 0. By continuity, then, in circling the contour, the vector χ(x) will also rotate by
2 π.
Thus, there will be a point within each such contour at which ϕ(x) = 0. This implies
that somewhere in space there exists a curve — either closed or infinite — upon which
ϕ(x) = 0. The existence of such a curve is energetically unfavorable, since ϕ ≪ ϕ0
nearby and the gradient of the field ϕ is also nonzero. However, topological considerations
indicate that such a curve, once having been produced during a phase transition, cannot
break; only if it is closed can it shrink to a point and disappear. The curve ϕ(x) = 0
owes its topological stability to the fact that as one goes around this curve, the vector
χ(x) executes either no full rotations, or one, two, or three, but there is no continuous
transformation between the corresponding distributions of the field χ (in traveling along
the closed contour, and returning to the same point x, the vector χ(x) cannot make 0.99
full rotations in (χ, χ∗ ) space). Such curves, together with their surrounding regions of
inhomogeneous field χ(x), are called strings.
Similar configurations of the field χ can also arise in the Higgs model itself. In that
case, however, everywhere except on the curve ϕ(x) = 0 one can carry out a gauge
transformation like (1.1.16) and transform away the field ζ(x). However, this leads to the
appearance of a field Aµ (x) 6= 0 within the string, which contains a quantum of magnetic
flux H = ∇ × A. Such strings are entirely analogous to Abrikosov filaments in the theory
of superconductivity [190]. Just as before, it is impossible to break such a string, in the
present case by virtue of the conservation of magnetic flux. In order to distinguish such
strings from those devoid of gauge fields, the latter are sometimes called global strings
(their existence being related to global symmetry breaking).
Inasmuch as the directions of the isotopic vectors χ(x) are practically uncorrelated
in every region of size O(T−1 c ) immediately after the phase transition, strings initially
look like Brownian trajectories with “straight” segments whose characteristic length is
O(T−1c ). Gradually straightening out, these strings then accelerate as a result of their
tension, and start to move at close to the speed of light. The end result is that small
closed strings (with sizes less than O(t)) start to collapse, intersect, radiate their energy in
 102

the form of gravitational waves, and finally disappear. Very long strings, with sizes of the
order of the distance to the horizon ∼ t, bacome almost straight. If, as appears possible,
intersecting strings have a non-negligible probability of coalescing, thereby forming small
closed strings, the number of long, straight strings remaining within the horizon ought to
decrease to a value of order unity.
Let α be the energy density of a string of unit length. In theories with coupling
constants of order unity, α ∼ ϕ20 . The mass of a string inside the horizon is of order
δM ∼ α t ∼ ϕ20 t, while according to (1.3.20) the total mass of matter inside the horizon is
M ∼ 10 M2P t. This means that due to the evolution of strings, the universe will eventually
contain density inhomogeneities [192, 81]

δρ δM α ϕ2
∼ ∼ 10 2 ∼ 10 02 . (6.2.3)
ρ M MP MP

δρ
For α ∼ 10−6M2P , ϕ0 ∼ 1016 GeV, we obtain ∼ 10−5 , as required for galaxy formation.
ρ
In deriving this estimate, we have assumed that small closed strings rapidly (in a time
of order t) radiate their energy and vanish. Actually, this will only happen if the value
of α is large enough. More refined estimates [193] lead to values of α similar to those
obtained above,
α ∼ 2 · 10−6 M2P .
Notice that the typical mass scale and the value of ϕ0 that appear here are close to those
associated with symmetry breaking in grand unified theories. Such strings can exist in
some grand unified theories. Unfortunately, it is far from simple to arrange for such heavy
strings to be created after inflation, since the temperature of the universe after inflation is
typically much lower than ϕ0 ∼ 1016 GeV, and the phase transition which leads to heavy
string formation typically does not occur. Some possibilities for the formation of heavy
strings in the inflationary universe scenario will be discussed in the next chapter.
Now let us look at another important class of topologically stable objects which might
be formed at the time of phase transitions. To this end, we analyze symmetry breaking
in the O(3)-symmetric model of the scalar field ϕa , a = 1, 2, 3:

1 µ2 a 2 λ
L= (∂µ ϕa )2 + (ϕ ) − [(ϕa )2 ]2 . (6.2.4)
2 2 4
Symmetry breaking occurs in this model as a result of the appearance of the scalar field
µ
ϕa , with absolute value ϕ0 equal to √ , but with arbitrary direction in isotopic space
λ
(ϕ1 , ϕ2 , ϕ3 ). At the time of the phase transition, domains can appear such that the vector
ϕa can look either “out of” or “into” each domain (in isotopic space) at all points on its
surface. One example is the so-called “hedgehog” distribution shown in Fig. 6.2,
xa
ϕa (x) = ϕ0 f (r) , (6.2.5)
r
 103

Figure 6.2: Distribution of the field ϕa (6.2.5) about the center of a hedgehog (global
monopole).

µ √
where ϕ0 = √ , r = x2 , and f (r) is some function that tends to ±1 for r ≫ µ−1 , and
λ
tends to zero as r → 0 (the latter condition derives from the continuity of to the function
ϕa (x)). Such a distribution is a solution of equations of motion in the theory (6.2.4) (for a
specific choice of function f (r) with the indicated properties), and this solution turns out
to be topologically stable for the same reason as do the global strings considered above.
At large r, the main contribution to the hedgehog energy comes from gradient terms
xa
arising from the change in direction of the unit vector at different points,
r
1 3 ϕ20
ρ≈ (∂i ϕ)2 = , (6.2.6)
2 2 r2
whereupon that part of the hedgehog energy contained within a sphere of radius r centered
at x = 0 is
E(r) = 6 π ϕ20 r . (6.2.7)
In infinite space, the total hedgehog energy thus goes to infinity (as r). That is why
the hedgehog solution (6.2.5), discovered more than ten years ago in the same paper as
monopoles [83], failed until fairly recently to elicit much interest in and of itself.
When phase transitions take place in an expanding universe, however, hedgehogs can
most certainly be created. The theory of hedgehog formation is similar to the theory of
string creation, and in fact the first estimates of the number of monopoles created during
a phase transition [40] were based implicitly on an analysis of hedgehog production. An
investigation of this problem shows that rather than being created singly, hedgehogs
are typically created in hedgehog-antihedgehog pairs (corresponding to f (r) = ±1 for
r ≫ m−1 in (6.2.5)). At large distances, such a pair exerts a mutually compensatory
influence on the field ϕ, and instead of the infinite energy of the individual hedgehogs, we
obtain the energy of the pair, which is proportional to their mutual separation r (6.2.7).
This is the simplest example of a realization of the idea of confinement that we know of.
 104

Subsequent evolution of a hedgehog-antihedgehog molecule depends strongly on hedge-

hog interactions with matter. In a hot universe, such a molecule is initially not very large,
2 −1
r< ∼ 10 Tc . Actually, though, according to the results of the previous section, domains
filled with the homogeneous field ϕ have characteristic sizes of order 10 T−1c (see (6.1.4)).
Simple combinatoric arguments indicate that in a region containing 10 –103 such domains
2

with uncorrelated values of ϕa , one will surely find at least one hedgehog, thereby yielding
the preceding estimate.
If the fields ϕa interact weakly with matter, hedgehogs and antihedgehogs quickly
approach one another, start executing oscillatory motion, radiate Goldstone bosons and
gravitational waves, approach still closer, and finally annihilate, radiating their energy in
the same way as do closed (global) strings. But if hedgehog motion is strongly damped by
matter, the annihilation process can take much longer. We shall return to the discussion of
possible cosmological effects associated with hedgehogs when we consider the production
of density inhomogeneities in the inflationary universe scenario.
If we supplement the theory (6.2.4) with O(3)-symmetric Yang–Mills fields with a
coupling constant e, the resultant theory will also have a solution of the equations of
motion like (6.2.5) for the field ϕa , but classical Yang–Mills fields will show up as well.
By a gauge transformation of the fields ϕa and Aaµ , we can “comb out” the hedgehog, i.e.,
send the fields ϕa off in one direction (for example, ϕa ∼ x3 δ3a ) everywhere except along
some infinitely thin filament emanating from the point x = 0. Far from the point x = 0,
then, the vector fields Aµ1,2 acquire a mass mA = e ϕ0 , while the vector field A3µ remains
massless. The most important feature of the resulting configuration of the fields ϕa and
Aaµ is then the presence of a magnetic field H = ∇ × A3 which falls off far from the center,

1 x
H= . (6.2.8)
e r3
Hence, this theory gives rise to particles analogous to the Dirac monopole (’t Hooft–
Poljakov monopoles) with magnetic charge
4π
g= , (6.2.9)
e
and these particles have an extremely high mass,
!
λ 4 π mA c mA
M=c = , (6.2.10)
e2 e2 α

e2
!
λ
where α = , and c is a quantity approximately equal to unity: (c(0) = 1,
4π e2
c(0.5) = 1.42, c(10) = 1.44).
In contrast to hedgehogs (6.2.5), ’t Hooft–Polyakov monopoles ought to exist in all
grand unified theories, in which the weak, strong, and electromagnetic interactions prior
to symmetry breaking are described by a single theory with a simple symmetry group
(SU(5), O(10), E6 , . . . ). Just as for hedgehogs, monopoles are produced during phase
 105

transitions, separated from each other by distance of order 102 T−1 c . Their initial density
−6
nM at the phase transition epoch was thereby 10 times the photon density nγ . Zeldovich
and Khlopov’s study of the monopole-antimonopole annihilation rate [40] has shown that
nM
annihilation proceeds very slowly, so that at present we should find ∼ 10−9 –10−10 , i.e.,
nγ
nM ≈ nB , where nB is the baryon (proton and neutron) density. The present density ρB of
baryon matter in the universe differs from the critical density by no more than one or two
orders of magnitude, ρB ∼ 10−29 g/cm3 . In grand unified theories, according to (6.2.10),
monopoles should have a mass of 102 MX ∼ 1016 –1017 GeV; that is, 1016 –1017 times the
mass of the proton. But that would mean, if we believe the estimate nM ≈ nB , that the
density of matter in the universe exceeds the critical value by 16 orders of magnitude.
Such a universe would already have collapsed long ago!
Even more stringent limits are placed on the allowable present-day density by the
existence of the galactic magnetic field [194], and by theoretical estimates of pulsar lumi-
nosity [195] due to monopole catalysis of proton decay [196]. These constraints lead one
nM
to conclude that at present, most likely < 10−25 –10−30 . Such an enormous disparity
nB ∼
between the observational constraints on the density of monopoles in the universe and
the theoretical predictions have led us to the brink of a crisis: modern elementary par-
ticle theory is in direct conflict with the hot universe theory. If we are to get rid of this
contradiction, we have three options:
a) renounce grand unified theories;
b) find conditions under which monopole annihilation proceeds much more efficiently;
c) renounce the standard hot universe theory.
At the end of the 1970’s, the first choice literally amounted to blasphemy. Later
on, after the advent of more complicated theories based on supergravity and superstring
theory, the general attitude toward grand unified theories began to change. But for the
most part, rather than helping to solve the primordial monopole problem, the new theories
engender fresh conflicts with the hot universe theory that are just as serious; see Section
1.5.
The second possibility has so far not been carried through to completion. The ba-
sic conclusions of the theory of monopole annihilation proposed in [40] have since been
confirmed by many authors. On the other hand, it has been argued [173] that nonpertur-
bative effects in a high-temperature Yang–Mills gas can lead to monopole confinement,
accelerating the annihilation process considerably.
The basic idea here is that far from a monopole, its field is effectively Abelian,
H = ∇ × Aa · δ3a . Such a field satisfies Gauss’ theorem identically, ∇ · H = 0, so
a

its flux is conserved. However, if the Yang–Mills fields in a hot plasma acquire an effec-
tive magnetic mass mA ∼ e2 T (see Section 3.3), then the monopole magnetic field will
be able to penetrate the medium only out to a distance m−1 A . The only way to make
this condition compatible with the magnetic field version of Gauss’ theorem is to invoke
filaments of thickness ∆l ∼ m−1 A emanating from the monopoles and incorporating their
entire magnetic field. But this is exactly how the magnetic field of a monopole embedded
 106

Figure 6.3: Magnetic field configuration for a monopole-antimonopole pair embedded in

a superconductor.

in a superconductor behaves (and for the same reason): Abrikosov filaments (strings) of
the magnetic field come into being between monopoles and antimonopoles [190]; see Fig.
6.2. Since the energy of each such string is proportional to its length, monopoles in a
superconductor ought to be found in a confinement phase [197]. If the analogous phe-
nomenon comes into play in the hot Yang–Mills gas, then the monopoles there should be
bound to antimonopoles by strings of thickness ∆l ∼ (e2 T)−1 . Monopole-antimonopole
pairs will therefore annihilate much more rapidly than when they are bound solely by
conventional attractive Coulomb forces.
Unfortunately, we still have too imperfect an understanding of the thermodynamics of
the Yang–Mills gas to be able to confirm or refute the existence of monopole confinement
in a hot plasma. Nonperturbative analysis of this problem using Monte Carlo lattice
simulations [198, 199] is not particularly informative, since the use of the lattice gives rise
to fictitious light monopoles whose mass is of the order of the reciprocal lattice spacing
a−1 . These fictitious monopoles act to screen out the mutual interaction of ’t Hooft–
Polyakov monopoles during the Monte Carlo simulations; they are difficult to get rid of
with presently available computing capabilities.
Besides the monopole confinement mechanism discussed above, there is another that
is even simpler [200]. Specifically, it is well known that in addition to not being able
to penetrate a superconductor, a magnetic field cannot penetrate the bulk of a perfect
conductor either (if the field was not present in the conductor from the very start), the
reason being that induced currents cancel the external magnetic field. The conductivity of
the Yang–Mills plasma is extremely high, and that is why when monopoles appear during
a phase transition, their magnetic field does not make its appearance in the medium right
away. The entire magnetic field must first be concentrated into some string joining a
monopole and antimonopole, as in Fig. 6.2 (due to conservation of total magnetic flux,
induced currents cannot cancel the entire magnetic flux, which passes along the filament).
The string will gradually thicken, and the field will adopt its usual Coulomb configuration.
If, however, the rate at which the thickness of the string grows is small compared with
the rate at which the monopoles separate from one another due to the expansion of the
universe, the field distribution will remain one-dimensional for a long time; in other words,
a confinement regime will prevail once again. Our estimates show that such a regime is
actually possible in grand unified theories.
A preliminary analysis of the annihilation of monopoles in a confinement phase indicate
 107

that the monopole density at the present epoch may be 10–20 orders of magnitude lower
than was first thought. A complete solution of this problem is exceedingly difficult,
however, and it is not clear whether monopole confinement will provide a way to reconcile
theoretical estimates of their density with the most stringent experimental limits, namely
those based on the existence of galactic magnetic fields and the observed lack of strong
X-ray emission from pulsars.
The theory of the interaction of monopoles with matter may yet harbor even more
surprises. But even if a way were found to solve the primordial monopole problem within
the framework of the standard hot universe theory, it would be hard to overstate the value
of the contribution made by analysis of this problem to the development of contemporary
cosmology. It is precisely the numerous attempts to resolve this problem that have led
to wide-ranging discussions of the internal inconsistencies of the hot universe theory, and
to a recognition of the need to reexamine its foundations. These attempts served as an
impetus for the development of the inflationary universe scenario, and for the appearance
of new concepts relating both to the initial stages of the evolution of the observable part
of the universe and the global structure of the universe as a whole. We now turn to a
description of these concepts.
 7
General Principles of Inflationary
Cosmology

7.1 Introduction
In Chapter 1, we discussed the general structure of the inflationary universe scenario.
Recent developments have gone in three main directions:
a) studies of the basic features of the scenario and the revelation of its potential
capacity for a more accurate description of the observable part of the universe. These
studies deal basically with problems related to the production of density inhomogeneities
at the time of inflation, the reheating of the universe, and the generation of the post-
inflation baryon asymmetry, along with those predictions of the scenario that might be
tested by analysis of the available observational evidence;
b) the construction of realistic versions of the inflationary universe scenario based on
modern elementary particle theories;
c) studies of the global properties of space and time within the framework of quantum
cosmology, making use of the inflationary universe scenario.
The first of these avenues of research will be discussed in the present chapter. The
second will form the subject of Chapters 8 and 9, and the third, Chapter 10.
 109

7.2 The inflationary universe and de Sitter space

As we have already noted in Chapter 1, the main feature of the inflationary stage of
evolution of the universe is the slow variation (compared with the rate of expansion of the
universe) of the energy density ρ. In the limiting case ρ = const, the Einstein equation
(1.3.7) for a homogeneous universe has the de Sitter space (1.6.1)–(1.6.3) as its solution.
It is easy to see that when H t ≫ 1, the distinction between an open, closed, and flat
de Sitter space tends to vanish. Much less obvious is the fact that all three of the solutions
(1.6.1)–(1.6.3) actually describe the very same de Sitter space.
To facilitate an intuitive interpretation of a curved four-dimensional space, it is often
convenient to imagine it to be a curved four-dimensional hypersurface embedded in a
higher-dimensional space. De Sitter space is most easily represented as a hyperboloid

z02 − z12 − z22 − z32 − z42 = −H−2 (7.2.1)

in the five-dimensional Minkowski space (z0 , z1 , . . . , z4 ). In order to represent de Sitter

space as a flat Friedmann universe (1.3.2), (1.6.2), it suffices to consider a coordinate
system t, xi on the hyperboloid (7.2.1) defined by the relations
1
z0 = H−1 sinh H t + H eH t x2 ,
2
1
z4 = H cosh H t − H eH t x2 ,
−1
2
zi = eH t xi , i = 1, 2, 3 . (7.2.2)

This coordinate system spans the half of the hyperboloid with z0 + z4 > 0 (see Fig. 7.2),
and its metric takes the form

ds2 = dt2 − e2H t dx2 . (7.2.3)

De Sitter space looks like a closed Friedmann universe in the coordinate system
(t, χ, θ, ϕ) defined by

z0 = H−1 sinh H t
z1 = H−1 cosh H t cos χ ,
z2 = H−1 cosh H t sin χ cos θ ,
z3 = H−1 cosh H t sin χ sin θ cos ϕ ,
z4 = H−1 cosh H t sin χ sin θ sin ϕ . (7.2.4)

The metric then becomes

ds2 = dt2 − H−2 cosh2 H t [dχ2 + sin2 χ (dθ2 + sin2 θ dϕ2 )] . (7.2.5)

It is important to note that in contrast to the flat-universe metric (7.2.3) and the metric
for an open de Sitter space (which we will not write out here), the closed-universe metric
 110

z0

t = const

x = const

z4

z1

Figure 7.1: De Sitter space represented as a hyperboloid in five-dimensional space-time

(with two dimensions omitted). In the coordinates (7.2.2), three-dimensional space at
t = const is flat, expanding exponentially with increasing t — see (7.2.3). The coordinates
(7.2.2) span only half the hyperboloid.

(7.2.5) describes the entire hyperboloid. In the terminology of general relativity, one can
say that the closed de Sitter space, as distinct from the flat or open one, is geodesically
complete (see Fig. 7.2).
To gain some understanding of this situation, it is useful here to draw an analogy with
what happens near a black hole. In particular, the Schwarzschild metric does not provide
a description of events near the gravitational radius rg of the black hole, but there do
exist coordinate systems that enable one to describe what occurs within the black hole.
In the present instance, the analog of the Schwarzschild metric is the metric for a flat (or
open) de Sitter space. An even more complete analog is given by the static coordinates
(r, t, θ, ϕ):
√
z0 = H−2 − r 2 sinh H t ,
√
z1 = H−2 − r 2 cosh H t ,
z2 = r sin θ cos ϕ ,
z3 = r sin θ sin ϕ ,
z4 = r cos θ , 0 ≤ r ≤ H−1 . (7.2.6)

These coordinates span that part of the de Sitter space with z0 + z1 > 0, and the metric
 111

z0

t = const
x = const
z4

z1

Figure 7.2: De Sitter space, represented as a closed Friedmann universe with coordinates
(7.2.4), (7.2.5). These coordinates span the entire hyperboloid.

takes the form

ds2 = (1 − r 2 H2 ) dt2 − (1 − r 2 H2 )−1 dr 2 − r 2 (dθ2 + sin2 θ dϕ2 ) , (7.2.7)

resembling the form of the Schwarzschild metric

ds2 = (1 − rg r −1 ) dt2 − (1 − rg r −1 )−1 dr 2 − r 2 (dθ2 + sin2 θ dϕ2 ) , (7.2.8)

2M
where rg = , and M is the mass of the black hole. Equations (7.2.7) and (7.2.8)
M2P
demonstrate that de Sitter space in static coordinates comprises a region of radius H−1
that looks as if it were surrounded by a black hole. This result was provided with a
physical interpretation in Chapter 1 (see Eq. (1.4.14)) by introducing the concept of
the event horizon. The analogy between the properties of de Sitter space and those of a
black hole is a very important one for an understanding of many of the features of the
inflationary universe scenario, and it therefore merits further discussion.
It is well known that any perturbations of the black hole (7.2.8) are rapidly damped
out, and the only observable characteristic that remains is its mass (and its electric charge
and angular momentum if it is rotating). No information about physical processes occur-
ring inside a black hole leaves its surface (that is, the horizon located at r = rg ). This
set of statements (along with some qualifications and additions), is often known in the
literature as the theorem that a black hole has no hair; for example, see [119].
 112

The generalization of this “theorem” to de Sitter space [120, 121] reads that any
perturbation of the latter will be “forgotten” at an exponentially high rate; that is, after
a time t ≫ H−1 , the universe will become locally indistinguishable from a completely
homogeneous and isotropic de Sitter space. On the other hand, because of the existence of
an event horizon, all physical processes in a given region of de Sitter space are independent
of anything that happens at a distance greater than H−1 from that region.
The physical meaning of the first part of the theorem is especially transparent in the
coordinate system (7.2.3) (or (7.2.5) when t ≫ H−1 ): any perturbation of de Sitter space
that is entrained by the general cosmological expansion will be exponentially stretched.
Accordingly, spatial gradients of the metric, which characterize the local inhomogeneity
and anisotropy of the universe, are exponentially damped. This general statement, which
has been verified for a wide class of specific models [122], forms the basis for the solution
of the homogeneity and isotropy problems in the inflationary universe [54–56].
The second part of the theorem means that if the initial size of an inflationary region
exceeds the distance to the horizon (r > H−1 ), then no events outside that region can
hinder its inflation, since no information about those events can ever reach it. The in-
difference of inflationary regions to what goes on about them might be characterized as
a sort of relatively harmless egoism: the growth of inflationary regions takes place basi-
cally by virtue of their inherent resources, rather than those of neighboring regions of the
universe. This kind of process (chaotic inflation) eventually leads to a universe with very
complex structure on enormous scales, but within any inflationary region, the universe
looks locally uniform to high accuracy. This circumstance plays an important role in
any discussion of the initial conditions required for the onset of inflationary behavior (see
Sections 1.7 and 9.1) or investigation of the global structure of the universe (Sections 1.8
and 10.2).
We shall return in the next section to the analogy between physical processes inside
a black hole and those in the inflationary universe, but here we should like to make one
more remark apropos of de Sitter space and its relation to the inflationary universe theory.
Many classic textbooks on general relativity theory treat de Sitter space as nothing
but the static space (7.2.7). As we have already pointed out, however, the space described
by the metric (7.2.7) is geodesically incomplete; that is, there exist geodesics that carry
one out of the space (7.2.7). In much the same way that an observer falling into a
black hole does not notice anything exceptional as he makes the final plunge through the
Schwarzschild sphere r = rg , so an observer in de Sitter space who is located at some
initial point r = r0 < H−1 emerges from the region described by the coordinates (7.2.7)
after a definite proper time interval (as measured by his own clocks). (While this is going
on, a stationary observer located at r = ∞ in the metric (7.2.8) or at t = 0 in the metric
(7.2.7) will never expect to see his friend disappear beyond the horizon, but he will receive
less and less information.) At the same time, the geodesically complete space (7.2.5) is
non-static.
In the absence of observers, matter, or even test particles, this lack of stationarity is
a “thing unto itself,” since the invariant characteristics of de Sitter space itself that are
associated with the curvature tensor are time-independent. Thus, for example, the scalar
 113

curvature of de Sitter space is

R = 12 H2 = const . (7.2.9)
Therefore, if the inflationary universe were simply an empty de Sitter space, it would be
difficult to speak of its expansion. It would always be possible to find a coordinate system
in which de Sitter space looked, for example, as if it were contracting, or as if it had a size
∼ H−1 (Eqs. (7.2.5), (7.2.7)). But in the inflationary universe, the de Sitter invariance
is either spontaneously broken (due to the decay of the initial de Sitter vacuum), or is
broken on account of an initial disparity between the actual universe and de Sitter space.
In particular, the energy-momentum tensor Tµν in the chaotic inflation scenario, even
though it is close to V(ϕ) gµν , is never exactly equal to the latter, and in the last stages of
1
inflation, the relative magnitude of the field kinetic energy ϕ̇2 becomes large compared
2
to V(ϕ), and the difference between Tµν and V(ϕ) gµν becomes significant. The distinction
between static de Sitter space and the inflationary universe becomes especially clear at
δρ
the quantum level, when one analyzes density inhomogeneities that arise at the time
ρ
of inflation. As we will show in Section 7.5, by the end of inflation, these inhomogeneities
δρ H2
grow to ∼ . Thus, if the field ϕ were constant and the inflating universe were
ρ ϕ̇
indistinguishable from de Sitter space, then after inflation ended our universe would be
highly inhomogeneous. In other words, a correct treatment of the inflationary universe
requires that we not only take its similarities to de Sitter space into account, but its
differences as well, especially in the latest stages of inflation, when the structure of the
observable part of the universe was formed.

7.3 Quantum fluctuations in the inflationary universe

The analogy between a black hole and de Sitter space is also useful in studying quantum
effects in the inflationary universe. It is well known, for example, that black holes evap-
M2P 1
orate, emitting radiation at the Hawking temperature TH = = , where M is
8πM 4 π rg
the mass of the black hole [119]. A similar phenomenon exists in de Sitter space, where
H
an observer will feel as if he is in a thermal bath at a temperature TH = . Formally, we
2π
can see this making the substitution t → i τ in Eq. (7.2.5) in order to make the transition
to the Euclidean formulation of quantum field theory in de Sitter space. The metric then
becomes that of a four-sphere S4 ,
−ds2 = dτ 2 + H−2 cos2 H τ [dχ2 + sin2 χ (dθ2 + sin2 θ dϕ2 )] . (7.3.1)
2π
Bose fields on the sphere are periodic in τ with period , which is equivalent to con-
H
H
sidering quantum statistics at a temperature TH = [201]. Physically, the appearance
2π
 114

of a temperature TH in de Sitter space (as is also the case for a black hole) is related
to the necessity of averaging over states beyond the event horizon [119, 120]. However,
the “temperature” of de Sitter space is highly unusual, in that the Euclidean sphere S4 is
periodic in all four directions, so the vacuum fluctuation spectrum turns out to be quite
unlike the usual spectrum of thermal fluctuations.
Averages like hϕ(x) ϕ(y)i and hϕ(x)2 i will play a particularly important role in our
investigation. In Minkowski space at a finite temperature T
∞
T d3 k
hϕ(x)2 i =
X
, (7.3.2)
(2 π)3 n=−∞ (2 π n T)2 + k2 + m2
which reduces to Eq. (3.1.7) for hϕ2 i after summing over n. In S4 -space, all integrations
are replaced by summations over ni , i = 1, 2, 3, 4, and the temperature is replaced
H
by the quantity . A term with ni = 0 is especially important in summing over ni ,
2π
since it makes the leading contribution to hϕ2 i as m2 → 0. It is readily shown that this
H2
contribution will be proportional to 2 ; for m2 ≪ H2 , the corresponding calculation gives
m
3 H4
hϕ2 i = (7.3.3)
8 π 2 m2
(a result first obtained by a different method [202, 126–128]). The pathological behavior of
hϕ2 i as m2 → 0 is noteworthy. Formally, it occurs because now instead of one summation,
we have four, and the corresponding infrared divergences of scalar field theory in de Sitter
space are found to be three orders of magnitude stronger than in quantum statistics.1 It
will be very important to understand the physical basis of such a strange result.
To this end, one quantizes the massless scalar field ϕ in de Sitter space in the coor-
dinates (7.2.3) in much the same way as in Minkowski space [202, 126–128]. The scalar
field operator ϕ(x) can be represented in the form
Z
ϕ(x, t) = (2 π)−3/2 d3 p [a+
p ψp (t) e
ipx
+ a− ∗
p ψp (t) e
−i p x
], (7.3.4)

where according to (1.7.13), ψp (t) satisfies the equation

ψ̈p (t) + 3 Hψ̇p (t) + p2 e−2 H t ψp (t) = 0 . (7.3.5)

1 √
In Minkowski space, the function √ e−ipt takes on the role of ψp (t), where p = p2 ;
2p
see (1.1.3). In de Sitter space (7.2.3), the general solution of (7.3.5) takes the form
√
π (1) (2)
ψp (t) = H η 3/2 [C1 (p) H3/2 (p η) + C2 (p) H3/2 (p η)] , (7.3.6)
2
1
Note that the vector or spinor field theory, sums over ni contain no terms that are singular in the limit
m→0
 115

(i)
where η = −H−1 e−H t is the conformal time, and the H3/2 are Hankel functions:
s
2 −i x 1
 
(2) (1)
H3/2 (x) = [H3/2 (x)]∗ =− e 1+ . (7.3.7)
πx ix
Quantization in de Sitter space and Minkowski space should be identical in the high-
frequency limit, i.e., C1 (p) → 0, C2 (p) → −1 as p → ∞. In particular, this condition is
satisfied2 for C1 ≡ 0, C2 ≡ −1. In that case,
iH p −H t i p −H t
   
ψp (t) = √ 1+ e exp e . (7.3.8)
p 2p iH H

Notice that at sufficiently large t (when p e−H t < H), ψp (t) ceases to oscillate, and becomes
iH
equal to √ .
p 2p
The quantity hϕ2 i may be simply expressed in terms of ψp :

e−2H t H2
!
2 1 Z 2 3 1 Z
hϕ i = |ψp | d p = + 3 d3 p . (7.3.9)
(2 π)3 (2 π)3 2p 2p

The physical meaning of this result becomes clear when one transforms from the conformal
momentum p, which is time-independent, to the conventional physical momentum k =
p e−H t , which decreases as the universe expands:

d3 k H2
!
1 1
Z
2
hϕ i = + . (7.3.10)
(2 π)3 k 2 2 k2

The first term is the usual contribution from vacuum fluctuations in Minkowski space (for
H = 0; see (2.1.6), (2.1.7)). This contribution can be eliminated by renormalization, as in
the theory of phase transitions (see (3.1.6)). The second term, however, is directly related
to inflation. Looked at from the standpoint of quantization in Minkowski space, this term
arises because of the fact that de Sitter space, apart from the usual quantum fluctuations
that are present when H = 0, also contains ϕ-particles with occupation numbers

H2
nk = . (7.3.11)
2 k2
It can be seen from (7.3.10) that the contribution to hϕ2 i from long-wave fluctuations of
the ϕ field diverges, and that is why the value of hϕ2 i in Eq. (7.3.3) becomes infinite as
m2 → 0.
However, the value of hϕ2 i for a massless field ϕ is infinite only in de Sitter space
that exists forever, and not in the inflationary universe, which expands exponentially (or
quasiexponentially) starting at some time t = 0 (for example, when the density of the
2
It is important that if the inflationary stage is long enough, all physical results are independent of the
specific choice of functions C1 (p) and C2 (p) if C1 (p) → 0, C2 (p) → −1 as p → ∞.
 116

universe becomes smaller than the Planck density). Indeed, the spectrum of vacuum
fluctuations (7.3.10) differs from the fluctuation spectrum in Minkowski space only when
k< ∼ H. If the fluctuation spectrum before inflation has a cutoff at k < ∼ k0 ∼ T resulting
from high-temperature effects [127], or at k < ∼ 0k ∼ H due to an inflationary region of
−1
the universe having an initial size O(H ), then the spectrum will change at the time
of inflation, due to exponential growth in the wavelength of vacuum fluctuations. The
−H t
spectrum (7.3.10) will gradually be established, but only at momenta k > ∼ k0 e . There
will then be a cutoff in the integral (7.3.9). Restricting our attention to contributions made
by long-wave fluctuations with k < ∼ H, which are the only ones that will subsequently be
important for us, and assuming that k0 = O(H), we obtain

H2 H d3 k H2 0 k
Z Z
hϕ2 i ≈ = d ln
2 (2 π)3 He−H t k 4 π2 −H t H
2 Z Ht 3
H p H
≡ 2
d ln = t. (7.3.12)
4π 0 H 4 π2
As t → ∞, hϕ2 i tends to infinity in accordance with (7.3.3). A similar result is obtained
for a massive scalar field ϕ. In that case, long-wave fluctuations with m2 ≪ H2 behave as

3 H4 2 m2
" !#
2
hϕ i = 1 − exp − t . (7.3.13)
8 π 2 m2 3H

3H 2
When t <∼ m2 , hϕ i grows linearly, just as in the case of the massless field (7.3.12),
and it then tends to its asymptotic value (7.3.3).
Let us now try to provide an intuitive physical interpretation of these results. First,
note that the main contribution to hϕ2 i (7.3.12) comes from integrating over exponentially
small k (with k ∼ H exp(−H t)). The corresponding occupation numbers nk (7.3.11) are
then exponentially large. For large l = |x − y| eH t , the correlation function hϕ(x) ϕ(y)i
for the massless field ϕ is [203]
1
 
2
hϕ(x, t) ϕ(y, t)i ≈ hϕ (x, t)i 1 − ln H l . (7.3.14)
Ht
This means that the magnitudes of the fields ϕ(x) and ϕ(y) will be highly correlated
out to exponentially large separations l ∼ H−1 exp(H t). By all these criteria, long-wave
quantum fluctuations of the field ϕ with k ≪ H−1 behave like a weakly inhomogeneous
(quasi)classical field ϕ generated during the inflationary stage; see the discussion of this
point in Section 2.1.
Analogous results also hold for a massive field with m2 ≪ !H2 . There, the principal con-
2
3H
tribution to hϕ2 i comes from modes with k ∼ H exp − , and the correlation length
2 m2
is of order
2
!
3 H
H−1 exp ; see Fig. 7.3.
2 m2
 117

ϕ
l

0 x

Figure 7.3: Distribution of the quasiclassical field ϕ generated at the time of inflation.
H √
For a massless field, dispersion ∆ is equal to H t, and a typical correlation length l
2π
is equal to H−1 exp(H t). For a massive field with m ≪ H, an equilibrium ! distribution is
H H2 −1 3 H2
established in a time ∆t <∼ m2 , having ∆ ∼ m and l ∼ H exp 2 m2 .

An important remark is in order here. In constructing theories of particle creation

in an expanding universe, elementary particle theorists have had to come to grips with
the fact that distinguishing real particles from vacuum fluctuations in the general theory
of relativity is a rather ambiguous problem [74]. What we have encountered here in our
example is a similar situation. Specifically, from the standpoint of quantization in the
coordinate system (7.2.3), long-wave fluctuations with H e−H t < ∼ k <∼ H correspond to
< < Ht
momenta H ∼ p ∼ H e . The corresponding occupation numbers in p-space show no
exponential rise with time whatsoever. The correlation between ϕ(x) and ϕ(y) at large
|x − y| is negligible. Thus, from the standpoint of quantization in de Sitter space (7.2.3),
we are dealing with quantum fluctuations. But from the standpoint of occupation numbers
at physical momenta k = p exp(−H t) and the correlation at large physical separations
l = |x − y| eH t , we are dealing with a quasiclassical weakly inhomogeneous field ϕ.
The difference inquestion is quite evident when we compare the functions ψp (t) (7.3.8)
3

and ψk (t) = ψp exp H t , the square of which also gives the spectrum (7.3.10) in terms
2
of the physical momentum k:
1 H
 
ik
ψk (t) = − √ e− H 1+ . (7.3.15)
2 ik
1
When k ≫ H, we are dealing with a field that oscillates at constant amplitude √ . But
2
in the course of time, when the magnitude of k ∼ p e−H t (p = const) falls below H,
oscillations will cease, and the amplitude of the field distribution for ψk (t), which has had
 118

its phase frozen in, begins to grow exponentially,

iH iH
ψk (t) = √ ∼ √ eH t , (7.3.16)
2k 2p
which is just the reason for the appearance of exponentially large occupation numbers.
We have already encountered this phenomenon in discussing the problem of Bose con-
densation and symmetry breaking in field theory. This is exactly the way in which the
exponential instability involving the creation of the classical Higgs field develops; see Eq.
(1.1.6). The difference here is that for symmetry breaking in Minkowski space, it is the
mode with vanishing momentum k that grows most rapidly. In the inflationary universe,
the momentum k at any of the modes falls off exponentially. This leads to an almost iden-
tical growth of modes with different initial momenta k, as a result of which the classical
field ϕ becomes inhomogeneous, although this inhomogeneity becomes significant only at
exponentially large distances l ∼ H−1 exp(H t); see (7.3.14). Another important differ-
ence between the phenomenon at hand and spontaneous symmetry breaking in Minkowski
space is that the production of a classical field ϕ in de Sitter space is an induced phe-
nomenon. The growth of long-wave perturbations of the field ϕ occurs even when it is
energetically unfavorable, as for instance when m2 > 0 (but only when m2 ≪ H2 ).
The process for generating a classical scalar field ϕ(x) in the inflationary universe can
be interpreted to be the result of the Brownian motion of the field ϕ induced by the con-
version of quantum fluctuations of that field into a quasiclassical field ϕ(x). For any given
mode with fixed p, this conversion occurs whenever the physical momentum k ∼ p e−H t
becomes comparable to H. A “freezing” of the amplitude of the field ψp (t) then occurs;
see (7.3.8). Due to a phase mismatch ei p x , waves with different momenta contribute to
the classical field ϕ(x) with different signs, and this also shows up in Eq. (7.3.9), which
characterizes the variance in the random distribution of the field that arises at the time
of inflation. As in the standard diffusion problem for a particle undergoing Brownian
motion, the mean squared particle distance from the origin is directly proportional to the
duration of the process (7.3.12).
At any given point, the diffusion of the field ϕ can conveniently be described by the
probability distribution Pc (ϕ, t) to find the field ϕ at that point and at a given instant
of time t. The subscript c here serves to indicate the fact that this distribution, as can
readily be shown, also corresponds to the fraction of the original coordinate volume d3 x
¯
(7.2.3) filled by the field ϕ at time t. The evolution of the distribution of the massless
field ϕ in the inflationary universe can be found by solving the diffusion equation [204,
134, 135]:
∂Pc (ϕ, t) ∂ 2 Pc (ϕ, t)
=D . (7.3.17)
∂t ∂ϕ2
To determine the diffusion coefficient D in (7.3.17), we take advantage of the fact that

2
Z
H3
hϕ i ≡ ϕ2 Pc (ϕ, t) dϕ = t.
4 π2
 119

Differentiating this relation with respect to t and using (7.3.17), we obtain

H3
D= .
8 π2
It is readily shown that the solution of (7.3.17) with initial condition Pc (ϕ, 0) = δ(ϕ) is a
Gaussian distribution s
2 π 2 ϕ2
!
2π
Pc (ϕ, t) = exp − 3 , (7.3.18)
H3 t H t
2 2 H3 t
with dispersion squared ∆ = hϕ i = (7.3.12).
4 π2
When we consider the production of a massive classical scalar field with mass |m2 | ≪
H2 , the diffusion coefficient D, which is related to the rate at which quantum fluctuations
with k > H are transferred to the range k < H, remains unchanged, since the contribution
to hϕ2 i from modes with k ∼ H does not depend on m for |m2 | ≪ H2 . For the same
reason, hϕ2 i of (7.3.13) grows in just the same way as for a massless field, as given by
(7.3.12). But subsequently, the long-wave classical field ϕ, which appears during the first
stages of the process, begins to decrease as a result of the slow roll down toward the point
ϕ = 0, in accordance with the classical equation of motion
dV
ϕ̈ + 3 H ϕ̇ = − = −m2 ϕ . (7.3.19)
dϕ

3 H4
This finally leads to stabilization of the quantity hϕ2 i at its limiting value (7.3.13).
8 π 2 m2
To describe this process, we must write the diffusion equation in a more general form [205]:

∂ 2 Pc
!
∂Pc ∂ dV
=D 2
+b Pc , (7.3.20)
∂t ∂ϕ ∂ϕ dϕ

H3
where as before D = and b is the mobility coefficient, defined by the equation
8 π2
dV
ϕ̇ = −b . Using (7.3.19) for the slowly varying field ϕ (ϕ̈ ≪ 3 H ϕ̇), we obtain
dϕ

H3 ∂ 2 Pc
!
∂Pc 1 ∂ dV
= 2 2
+ Pc . (7.3.21)
∂t 8 π ∂ϕ 3 H ∂ϕ dϕ

This equation was first derived by Starobinsky [134] by another, more rigorous method; a
more detailed derivation can be found in [186, 135, 132, 206]. Solution of this equation for
m2 2
the case V(ϕ) = ϕ + V(0) actually leads to the distribution Pc (ϕ, t) with dispersion
2
determined by (7.3.13). Solutions that are valid for a more general class of potentials
V(ϕ) will be discussed in the next section in connection with the problem of tunneling in
the inflationary universe.
 120

To conclude, we note that in deriving Eq. (7.3.21), it has been assumed that H is
independent of the field ϕ. More generally, Eq. (7.3.21) can be written in the form

∂2 H3 Pc
! !
∂Pc ∂ Pc dV
= + . (7.3.22)
∂t ∂ϕ2 8 π2 ∂ϕ 3 H dϕ

Strictly speaking, this equation also holds only when variations in the field ϕ are small
enough that the back reaction of inhomogeneities of the field on the metric is not too
large. Nevertheless, with the help of this equation, one can obtain important information
on the global structure of the universe; see Chapter 10.

7.4 Tunneling in the inflationary universe

The first versions of the inflationary universe scenario were based on the theory of decay
of a supercooled vacuum state ϕ = 0 due to tunneling with creation of bubbles of the field
ϕ at the time of inflation [53–55]. The theory of such processes in Minkowski space, which
is discussed in Chapter 4, turns out to be inapplicable to the most interesting situations,
where the curvature of the effective potential near its local minimum is small compared to
H2 . Coleman and De Luccia [207] have developed a Euclidean theory of tunneling in de
Sitter space, but the general applicability of this theory to the study of tunneling during
inflation was confirmed only very recently [208]. One of the main problems was that
according to [207] both the scalar field ϕ inside a bubble and the metric gµν (x) experience
a quantum jump. However, in certain situations, there is a barrier only in the direction of
change of the field ϕ. The analog of this problem is that of the motion of a particle in the
(x, y)-plane in a potential V(x, y) having the form of a barrier only in the x-direction. A
particle encountering the barrier in this situation tunnels through in the x-direction, but
it may continue to move undisturbed along its classical trajectory in the y-direction. To
investigate tunnelling under these circumstances, in general one cannot simply transform
to imaginary time (imaginary energy); instead, one must undertake an honest solution of
the Schrödinger equation for the wave function Ψ(x, y), allowing for the fact that some of
the components of the particle momentum may have an imaginary part.
Another problem was an ambiguity of interpretation of the results of Euclidean ap-
proach to tunneling. Let us consider e.g. a theory with the effective potential with a
2 
d V
small local minimum a ϕ = 0, such that m2 =  ≪ H2 ; tunneling in such a theory
dϕ2 ϕ=0
was studied by Hawking and Moss [121]. Their expression for the probability of tunneling
from the point ϕ = 0 through a barrier with a maximum at the point ϕ1 (Fig. 7.4) looks
like
3 M4P
" !#
1 1
P ∼ A exp − − , (7.4.1)
8 V(0) V(ϕ1 )
where A is some multiplicative factor with dimensionality m4 . Hawking and Moss assumed
in deriving this equation that by virtue of the “no hair” theorem for de Sitter space (see
 121

0 ϕ1 ϕ

Figure 7.4: The potential V(ϕ) used by Hawking and Moss to study tunneling.

Section 7.2), tunneling in an exponentially expanding de Sitter space (7.2.3) should occur
in just the same way it occurs in a closed space (7.2.5). Tunneling in the latter case is
most likely at the throat of the hyperboloid (i.e., at t = 0, a = H−1 ), while according
to [207], a description of that tunneling requires that we calculate the appropriate action
in the Euclidean version of the space (7.2.5), that is, on a sphere S4 of radius H−1 (ϕ).
Since our concern is with tunneling in which H(ϕ) increases (i.e. a decreases), which is
classically forbidden, the preceding argument against a Euclidean approach to this case
does not apply. A calculation of the action S on the sphere leads to the quantity
3 M4P
SE (ϕ) = − . (7.4.2)
8 V(ϕ)
Adhering to the ideology developed in the work of Coleman and De Luccia, Hawking and
Moss asserted that the probability of tunneling is proportional to exp(SE (0)−SE (ϕ)). This
also leads to Eq. (7.4.1), but the contribution to the action from the bubble walls was not
taken into account — in other words, they treated purely homogeneous tunneling suddenly
taking place over all space [121]. This result was later “confirmed” by numerous authors;
however, simultaneous tunneling throughout an entire exponentially large universe seems
quite unlikely.
In order to study this problem in more detail, a Hamiltonian approach to the theory
of tunneling at the time of inflation was developed, and has succeeded in showing that
the probability of homogeneous tunneling over an entire inflationary universe is actually
vanishingly small [186]. Hawking and Moss themselves later remarked that their result
should be interpreted not as the probability of homogeneous tunneling throughout the
entire universe, but as the probability of tunneling which looks homogeneous only on
some scale l > −1
∼ H [209]. They argued that bubble walls and other inhomogeneities
should have no effect on tunneling, due to the “no hair” theorem for de Sitter space (see
Section 7.2).
The validity of that argument, and in fact the overall applicability of the Euclidean
 122

approach to this problem, was open to doubt. Only much later was it learned that when
m2 ≪ H2 , the contribution to the Euclidean action from gradients of the field ϕ is small
[186] (in contrast to the situation in a Minkowski space, where this contribution is of
the same order as that of the potential energy of the field ϕ), and that tunneling in
this instance is effectively one-dimensional (basically occurring because of a change in the
scalar field). Equation (7.4.1) was thereby partially justified. But a genuine understanding
of the physical essence of this phenomenon was not achieved until an approach to the
theory of tunneling based on the diffusion equation (7.3.21) was developed [134, 135].
The fundamental idea is that for tunneling to occur, it suffices to have a bubble with
a field exceeding ϕ1 and radius
v
3 M2P
u
u
r > H−1 (ϕ) = t .
8 π V(ϕ1 )

Further evolution of the field ϕ inside this bubble does not depend on what goes on outside
it; in other words, the field will start to roll down to the absolute minimum of V(ϕ) with
ϕ > ϕ1 . It only remains to evaluate the probability that a region of this type will form
— but that is exactly the problem we studied in the previous section!
As we have already stated, the distribution Pc (ϕ, t) actually characterizes that fraction
of the original coordinate volume d3 x (7.2.3) which at time t contains the field ϕ; the latter
is homogeneous on a scale l > −1
∼ H . The problem of tunneling at the time of inflation
thereby reduces to the solution of the diffusion equation (7.3.21) with initial condition
Pc (ϕ, 0) = δ(ϕ).
At this point, we must distinguish between two possible regimes.
q H √
1) In the initial stage of this process, the dispersion hϕ2 i grows as H t (7.3.12).
2π
If at that stage it becomes larger in magnitude than ϕ1 , which characterizes the position
of the local maximum of V(ϕ), the process will proceed as if there were no barrier at all
[127]. In that event, diffusion will end when the field ϕ encounters a steep slope in V(ϕ),
where the rate of diffusive growth of the field drops below the rate of classical rolling.
Under typical conditions, the diffusion stage lasts for a time
4 π 2 ϕ21
t∼ , (7.4.3)
H3
and the typical shape of the regions within which the field ϕ exceeds some given value
(say ϕ1 ) is far from that of a spherical bubble.
q
2) If the dispersion stops growing when hϕ2 i ≪ ϕ1 , the distribution Pc (ϕ, t) will
∂Pc (ϕ, t)
become quasistationary. It should then be possible to find it by putting = 0 in
∂t
Eq. (7.3.21), or in the more general equation (7.3.22). To clarify the physical meaning of
solutions of these equations, it is convenient to rewrite Eq. (7.3.22) in the form
∂Pc ∂jc
= − , (7.4.4)
∂t ∂ϕ
 123
s
3 M2P 8 V2 ∂Pc
" !#
1 dV 4V
−jc = 4
+ Pc 1+ 4 . (7.4.5)
3 8 π V 3 MP ∂ϕ dϕ MP

Here we have introduced the probability current jc (ϕ, t) in (ϕ, t)-space [205], so that Eq.
(7.4.4) takes on the standard form of the continuity equation for the probability density
∂Pc
Pc (ϕ, t). Consideration of the standard condition = 0 amounts to examination of the
∂t
case in which the probability current is constant for all ϕ from −∞ to ∞. As a rule, there
are no reasonable initial conditions under which an unattenuated, nonvanishing diffusion
current jc = const 6= 0 arises between ϕ = −∞ and ϕ = +∞ (see [135], however).
Furthermore, the diffusion process itself is usually feasible only within certain limited
d2 V
ranges of variation of the field ϕ (namely, where ≪ H2 and V(ϕ) ≪ M4P ). Outside
dϕ2
these zones, the first (diffusion) term in (7.4.5) does not appear, and if the potential V(ϕ)
is an even function of ϕ, Eq. (7.4.5) implies that Pc must be an odd function of ϕ, which
is impossible, inasmuch as Pc (ϕ, t) ≥ 0. For all of these reasons, we will consider only the
case jc = 0 (in this regard, see also Chapter 10).
When jc = 0, and also when V(ϕ) ≪ M4P , Eq. (7.4.5) reverts to a very simple form,

∂ ln Pc 3 MP dV
=− , (7.4.6)
∂ϕ 8 V2 (ϕ) dϕ

whereupon
3 M4P
!
Pc = N exp , (7.4.7)
8 V(ϕ)
Z
where N is a normalizing factor such that Pc dϕ = 1. In the present instance, where the
q
rms deviation of the field is much less than the width of the potential well ( hϕ2 i ≪ ϕ1 ),
!
3 MP
the function exp has a clear-cut maximum at ϕ = 0, and therefore, to within
8 V(ϕ)
an unimportant subexponential factor,

3 M4P
!!
1 1
Pc (ϕ) = exp − − . (7.4.8)
8 V(0) V(ϕ)

According to (7.4.8), the probability that the field at a given point (or more precisely, in
a neighborhood of size l > −1
∼ H at a given point) will equal ϕ1 is given precisely by the
exponential term in the Hawking–Moss equation (7.4.1). This is not just a coincidence,
since the mean diffusion time from ϕ = 0 to ϕ = ϕ1 , that is, the mean time that
tunneling goes on at a given point, is in fact proportional to Pc (ϕ1 ). The corresponding
result in the theory of Brownian motion is well known [210]; for the present case, it
was derived in [134, 135]. Its physical meaning is most easily understood if we consider
motion along a Brownian trajectory at (approximately) constant speed (as happens here
when H(ϕ) ≈ const). The value of Pc (ϕ) indicates the number density of points on this
 124

trajectory at which the value of the field is equal to ϕ. This means that the mean time
τ to move from the point ϕ = 0 to the point ϕ = ϕ1 along a Brownian trajectory is
proportional to [Pc (ϕ)]−1 , and consequently the tunneling (diffusion) probability P per
unit time τ is proportional to Pc (ϕ).
Strictly speaking, the tunneling process is not stationary, but if the time required for
relaxation to the quasistationary state is much less than the time for tunneling, then
Eq. (7.4.8) will provide a good representation of the distribution Pc (ϕ) This condition is
satisfied if
3 M4P
!
1 1
− ≫1. (7.4.9)
8 V(0) V(ϕ1 )
q
One can readily show that this is equivalent to requiring that hϕ2 i ≪ ϕ1 . In the present
context, the probability of forming large nonspherical regions of a field ϕ > ϕ1 that are
bigger than H−1 (ϕ1 ) is much lower than that of forming spherical bubbles of the field ϕ.
As an example, consider the theory with effective potential

m2 2 λ 4
V(ϕ) = V(0) + ϕ − ϕ . (7.4.10)
2 4
m
For this theory, ϕ1 = √ , and Eq. (7.4.1) for V(ϕ1 ) − V(0) ≪ V(0) becomes
λ
3 M4P m4 2π 2
" # " 4 #
m

Pc ∼ exp − = exp − , (7.4.11)
32 λ V2 (0) 3λ H

while (7.4.9), together with the condition that m2 ≪ H2 , may be cast in the form
√ m2
λ< ≪1. (7.4.12)
H2
A more detailed study of the solutions of Eq. (7.3.22) makes it possible
q to obtain
expressions for the mean duration of tunneling which hold both for hϕ2 i ≫ ϕ1 and
q
hϕ2 i ≪ ϕ1 [135]. Most important for us here has been the elucidation of the general
properties of phase transitions at the time of inflation, a subject discussed in more detail
elsewhere [186]. One of the most surprising features of such phase transitions is the
possibility of diffusion from one local minimum of V(ϕ) to another with an increase
in energy density [211]. This effect and related ones are extremely important for an
understanding of the global structure of the universe. We shall return to this question in
Chapter 10.
Thus, with a stochastic approach, one can justify the Hawking–Moss equation (7.4.1)
[121] and confirm their interpretation of this equation [209]. On the other hand, this same
approach has provided a means of appreciating the limits of applicability of Eq. (7.4.1).
The “derivation” of this result given in [121] imposed no constraints on the form of the
potential V(ϕ), and it was not clear why tunneling should occur through the nearest
maximum of V(ϕ), rather than directly to its next minimum. The answer to this last
 125

question is obvious within the context of our present approach, and it is also clear that
Eq. (7.4.1) itself is only valid if the curvature of V(ϕ) is much less than H2 over the whole
domain of variation of ϕ from 0 to ϕ1 .
Another important observation to be made in studying the theory of tunneling in the
inflationary universe relates to the properties of the walls of bubbles of a new phase. In
Minkowski space, the total energy of a bubble of a new phase that is created from vacuum
is exactly zero. As the bubble grows, so does its negative energy, which is proportional to
4
its volume ∼ π r 3 ε, and which is related to the energy gain ε realized in the transition
3
to a new phase. At the same time (and the same rate), the positive bubble-wall energy
grows as 4 π r 2 σ(t), where σ is the surface energy density of the bubble. These two terms
cancel, which is only possible because the surface energy is also proportional to r, the
reason being that the speed of a wall approaches the speed of light, while its thickness
decreases. Thus, even if the thin-wall approximation were inappropriate in a description of
the bubble creation process, it could be usable in a description of its subsequent evolution
[212, 213]. Formally, this occurs because any bubble of the field ϕ created from vacuum
by O(4)-symmetric tunneling can be described by some function of the form

ϕ = ϕ(r2 − t2 ) ; (7.4.13)

see [180]. If this bubble has a characteristic initial size of r0 at t = 0, then the field at
large t will reach a value ϕ(0) at a distance

r02 r2
∆r = ≈ 0 (7.4.14)
2r 2t
from the bubble boundary (i.e., from the sphere on which ϕ(r2 − t2 ) = ϕ(r02 ) ≈ 0). With
time, then, the wall thickness quickly decreases.
In the inflationary universe, everything is completely different. The total energy of
the field ϕ within a bubble does not vanish, and is not conserved as the universe expands.
This is attributable to the same gravitational forces that drive the exponential growth of
the total energy of the scalar field at the time of inflation (E ∼ V(ϕ) a3 (t) ∼ V(ϕ) e3 H t ).
Tunneling results from the formation of perturbations δϕ(x) with wavelengths l > ∼H .
−1

All gradients of these perturbations are very small, and do not affect their evolution.
This is also the reason why in the final analysis the Hawking–Moss equation, neglecting
the contribution of boundary terms in the Euclidean action, is found to be correct. In
the study of bubbles engendered by the foregoing mechanism, therefore, the thin-wall
approximation is often inapplicable at any stage of bubble evolution. But if the regions
that are formed contain matter in many different phase states, then the domain walls
that appear between these regions in the late stages of inflation, or after inflation has
terminated, can actually become thin. The powerful methods developed in [212, 213] can
be utilized to investigate the structure of the universe in the vicinity of such regions.
 126

7.5 Quantum fluctuations and the generation of adiabatic density perturba-

tions
We now continue our study of perturbations of a scalar field with exponentially large
wavelength that come into being during an inflationary stage. From the standpoint of
quantization in the coordinate system (7.2.3), the wavelengths of these fluctuations do
not grow (p = const in Eq. (7.3.4)), and they differ but little from conventional vacuum
fluctuations. In particular, one can calculate the corrections to the energy-momentum
tensor gµν V(ϕ) that are associated with these fluctuations; in the stationary state (ϕ =
const), these are [202, 203]

1 2 2 3 H4
∆Tµν = hϕ i m gµν = gµν (7.5.1)
4 32 π 2
regardless of the mass of the field ϕ (for m2 ≪ H2 ). These corrections have a relativisti-
H
cally invariant form (despite the presence of the Hawking “temperature” TH = in de
2π
Sitter space).
But as we have already pointed out, from the point of view of a stationary observer
armed with measuring rods that do not stretch during inflation of the universe, fluctu-
ations of a scalar field that have wavelengths greater than the distance to the horizon
(k −1 >
∼ H ) look like a classical field δϕ that is weakly inhomogeneous on scales l >
−1
∼H .
−1

These fluctuations give rise to density inhomogeneities on an exponentially large scale.

During inflation, the magnitude of these inhomogeneities is

δρ ≈ V′ δϕ , (7.5.2)
dV
where V′ = . In the last stages of inflation, an ever increasing fraction of the field
dϕ
energy is tied up in the kinetic energy of the field ϕ, rather than in V(ϕ). This energy then
transforms into heat, and energy density inhomogeneities δρ lead to temperature inhomo-
geneities δT. The original density inhomogeneities (7.5.39) are thereby transformed into
hot-plasma density inhomogeneities, and then into cold-matter density inhomogeneities.
The corresponding density inhomogeneities result in so-called adiabatic perturbations of
the metric, in contrast to the isothermal perturbations associated with inhomogeneities
of the metric that arise at constant temperature.
The appearance of long-wave density (metric) perturbations is necessary for the subse-
quent formation of the large scale structure of the universe (galaxies, clusters of galaxies,
voids, and so on). Another possible mechanism for generating density perturbations is
related to the theory of cosmic strings produced during phase transitions in a hot uni-
verse. But it is very difficult to get along without an inflationary stage, and therefore
the prospect of obtaining inhomogeneities of the type required simply by virtue of quan-
tum effects at the time of inflation, without invoking any additional mechanisms, seems
especially interesting. The fact that the contribution to hϕ2 i (7.3.12) due to integration
k
over the fixed interval ∆ ln is independent of k leads to a flat spectrum δρ(k) (7.5.2)
H
 127

which does not depend on k (as the momentum varies on a logarithmic scale). This is
exactly the sort of spectrum suggested earlier by Harrison and Zeldovich [76] (see also
[214]) as the initial perturbation spectrum required for the subsequent formation of galax-
ies. If we normalize this spectrum so that δρ(k) denotes the contribution to δρ from all
k
perturbations per unit interval in ln , then the desired spectral amplitude should be
H
δρ(k)
∼ 10−4–10−5 (7.5.3)
ρ
on a galactic scale (lg ∼ 1022 cm at the present epoch; lg ∼ 10−5 cm at the instant that
inflation ended). Notice, however, that rather than referring to perturbations δρ at the
inflationary stage (7.5.2), condition (7.5.3) relates to their progeny at a later stage, after
ρ
reheating of the universe, when its equation of state becomes p = (or p = 0, when
3
cold nonrelativistic matter predominates). The question of how these perturbations actu-
ally relate to the initial perturbations (7.5.2) is a very complicated one. Some important
steps in the development of a theory of adiabatic density perturbations formed during the
exponential expansion stage of the universe are to be found in [101, 215–217]. The cor-
responding problem for the inflationary universe scenario was first solved by Mukhanov
and Chibisov [107] in their investigation of the Starobinsky model [52]. The quantity
δρ
for the new inflationary universe scenario was calculated by four other groups prac-
ρ
tically simultaneously [114]. The results obtained by all of these authors, using different
approaches, agreed to within a numerical factor C ∼ O(1):
δρ(k) H(ϕ) δϕ(k) 

=C . (7.5.4)
ρ ϕ̇ 
k∼H

δρ(k)
What this expression means is that in order to calculate on a logarithmic scale in k,
ρ
H[ϕ(t)]
one should calculate the value of the function at a time when the corresponding
ϕ̇(t)
wavelength k −1 is of the order of the distance to the horizon H−1 (that is, when the field
δϕ(k) becomes quasiclassical). For δϕ(k) here we can take the rms value defined by (see
(7.3.12))
H2 (ϕ) ln H +1 H2 (ϕ)
Z k
2 k
[δϕ(k)] = d ln = , (7.5.5)
4 π 2 ln Hk H 4 π2
or in other words
H(ϕ)
|δϕ(k)| = . (7.5.6)
2π
In the final analysis, these same results are found to be correct for the chaotic inflation
scenario as well [218].
It would be hard to overestimate the significance of the first papers on the density
perturbations produced during inflation [114]. However, the validity of some of the as-
sumptions made in these papers is not obvious. Furthermore, it was not entirely clear
 128

what the connection was between the density perturbations occurring during the infla-
tionary stage (7.5.2) and Eq. (7.5.4), and in fact the value of the parameter C in the latter
equation differed somewhat in the different papers. This situation engendered an exten-
sive literature on the problem; for a review, see [219]. In our opinion, a final clarification
of the situation was particularly facilitated by Mukhanov [218]. In what follows, we will
describe the basic ingredients of his work, and use his results to obtain an equation for
δρ
that will be valid for a large class of inflationary models.
ρ
Consider a region of the inflationary universe of initial size ∆l > −1
∼ H , containing a
i
sufficiently homogeneous field ϕ (∂i ϕ ∂ ϕ ≪ V(ϕ)). All initial inhomogeneities of this
field die out exponentially, and the total field in this region can therefore be put in the
form
ϕ(x, t) → ϕ(t) + δϕ(x, t) , (7.5.7)
where inhomogeneities δϕ(x, t) appear because of long-wave fluctuations that are gen-
erated with k < ∼ H. The leading contribution to δϕ(x, t) comes from fluctuations with
exponentially long wavelengths. Therefore, the main contribution to inhomogeneities
in the mean energy-momentum tensor Tνµ on the large scales that we are interested in
dV
come from terms like ∂0 ϕ · ∂0 [δϕ(x, t)] or δϕ(x, t) · , rather than from spatial gradi-
dϕ
ents (∂i [δϕ(x, t)])2 (the second of these is in fact the leading contribution at the time of
inflation; see (7.5.2)). This implies that δTνµ is diagonal to first order in δϕ. For pertur-
bations of this type, the corresponding perturbations of the metric in a flat universe can
be represented by [220]
ds2 = (1 + 2 Φ) dt2 − (1 − 2 Φ) a2 (t) dx2 . (7.5.8)
The function Φ(x, t) plays a role similar to that of the Newtonian potential used to
describe weak gravitational fields (compare the metric (7.5.8) with the Schwarzschild
metric (7.2.8)). The coordinate system (7.5.8) is more convenient for the investigation of
density perturbations than the more frequently used synchronous system [65], since after a
synchronous system is chosen through the condition δgi0 = 0, one still has the freedom to
change the coordinate system; this leads to the existence of two nonphysical perturbation
modes, which makes the calculations and their interpretation rather complicated and
ambiguous. However, these modes do not contribute to Φ(x, t). Density inhomogeneities
of the type we are considering, with wavelengths k −1 > H−1 , are related to the function
Φ(x, t) in a very simple way,
δρ
= −2 Φ . (7.5.9)
ρ
A more detailed discussion of the use of the relativistic potential Φ(x, t) may be found in
[220–222, 133]. Linearizing the Einstein equations and the equation for the field ϕ(x, t)
in terms of δϕ and Φ, one can obtain a system of differential equations for δϕ and Φ:
!  2 !
ȧ ϕ̈ 1 ä ȧ ȧ ϕ̈
Φ̈ + −2 Φ̇ − 2 ∆Φ + 2 − − Φ=0, (7.5.10)
a ϕ̇ a a a a ϕ̇
 129

1 4π
(a Φ)˙,β = 2 (ϕ̇ δϕ),β , (7.5.11)
a MP
ȧ 1 d2 V dV
δ ϕ̈ + 3 δ ϕ̇ − 2 ∆δϕ + 2
δϕ + 2 Φ − 4 ϕ̇ Φ̇ = 0 . (7.5.12)
a a dϕ dϕ
Here ϕ(t) and a(t) are the solutions of the unperturbed equations (see Section 1.7), and
a dot signifies differentiation with respect to time. Using one of the consequences of the
Einstein equations for a(t),
 ˙
ȧ 4π
= − 2 ϕ̇2 , (7.5.13)
a MP
Eq. (7.5.10) can be cast in the form
′′ (a′ /a2 ϕ′ )′′
u − ∆u − ′ 2 ′ u = 0 , (7.5.14)
a /a ϕ
a
where u = Φ, and primes in the (and only in this) equation stand for differentiation with
ϕ′
dt d2 V
Z
respect to the conformal time η = . In the long-wave limit (k ≪ H, k 2 ≪ ),
a(t) δϕ2
the solution of Eq. (7.5.14) can be put in the form
ȧ t
 Z 
Φ=C 1− 2 a dt , (7.5.15)
a 0
where C is some constant, and H tt ≫ 1. Thereupon, making use of (7.5.11), we obtain
1Z t
a dt .
δϕ = C ϕ̇ · (7.5.16)
a 0
The desired result follows from (7.5.15) and (7.5.16), namely the relationship between
long-wave fluctuations of the field ϕ, perturbations of the metric Φ, and density inhomo-
geneities [218]:
 
δρ a δϕ  ȧ t
 Z 

= −2 Φ = −2 Z
t
 1− a dt . (7.5.17)
ρ  ϕ̇  a2 0
a dt
0

The quantity in square brackets is the constant C of (7.5.15), the value of which can be
computed at any stage of inflation. This is most conveniently done when the wavelength
of a perturbation δϕ(k) is equal to the distance to the horizon, k ∼ H. The amplitude at
that time can be estimated with the help of Eq. (7.5.6).
We now make use of the preceding results to compare Eqs. (7.5.2) and (7.5.4), and we
δρ
apply these results to the calculation of in the simple models. One can easily verify
ρ
that during the stage of inflation, when Ḣ ≪ H2 , Ḧ ≪ H3 , and H t ≫ 1,
" !#!
a Ḣ Ḣ Ḧ
t = H(t) 1 + 2 1 + O , . (7.5.18)
H H2 H3
Z
a dt
0
 130

The expression in square brackets in (7.5.17) is thus equal to

δϕ
C = H(ϕ(t)) . (7.5.19)
ϕ̇
On the other hand, Eq. (7.5.18) implies that during the inflationary stage, the paren-
Ḣ
thesized expression in (7.5.17) equals 2 ≪ 1. In that event, it is readily verified using
H
(7.5.17) and (7.5.19) that density inhomogeneities during inflation are given by

δρ δV V′
≈ = δϕ , (7.5.20)
ρ V V
as one would find from (7.5.2), and not by! (7.5.4). The difference between (7.5.20) and
Ḣ
(7.5.4) lies in a small factor that is O ≪ 1.
H2
However, if the universe is hot (a ∼ t1/2 ) or cold (a ∼ t2/3 ), Eqs. (7.5.17) and (7.5.19)
lead to Eq. (7.5.4), with C = −4/3 and C = −6/5 for these two respective cases [218,
H δρ
220]. If for δϕ(k) we then take as in (7.5.6) in order to find the rms value of per
2π ρ
k
unit interval in ln , we obtain
H
δρ(k) [H(ϕ)]2 

=C . (7.5.21)
ρ 2 π ϕ̇ k∼H
Expressing ϕ̇ and H(ϕ) in terms of V(ϕ) during inflation, we find that at the stage when
cold matter predominates (when galaxy formation presumably started),
s
δρ(k) 48 2 π V3/2 

= (7.5.22)
ρ 5 3 dV k∼H(ϕ)
M3P
δϕ
(an irrelevant minus sign has been omitted from (7.5.22)). As an example of the use of
λ
this equation, one can obtain the amplitude of density inhomogeneities in the theory ϕ4
4
for the chaotic inflation scenario,
s
3 
δρ(k) 6 2πλ ϕ


=  . (7.5.23)
ρ 5 3 MP 
k∼H(ϕ)

δρ(k)
If we are to compare (7.5.23) with the value of on a galactic scale (lg ∼ 1022 cm)
ρ
or on the scale of the horizon (lH ∼ 1028 cm), we must follow the behavior of a wave
with momentum k during and after inflation. According to (1.7.25), a wave emitted
at some value of ϕ will, over the course of inflation, increase in wavelength by a factor
 131

ϕ2
!
exp π 2 . After reheating to a temperature TR and cooling to a temperature Tγ , where
MP
Tγ ∼ 3 K is the present-day temperature of the microwave background radiation, the
TR
universe will again typically expand by another factor of . Presuming that reheating
Tγ
MP
takes place immediately after the end of inflation (with ϕ ∼ ), TR will be of order
1/4
3
MP λ1/4
 
V ∼ MP . (The final results will be a very weak (logarithmic) function of
3 10
the duration of reheating and the magnitude of TR .) Thus, the present wavelength of a
perturbation produced at the moment when the scalar field had some value ϕ is of order

π ϕ2
!
−1TR
l(ϕ) ∼ H (ϕ) exp
Tγ M2P
2
! !
M P M P π ϕ
∼ M−1
P exp . (7.5.24)
ϕ λ1/4 Tγ M2P

Bearing in mind that 1 GeV corresponds approximately to 1013 K, MP ∼ 10−33 cm, and
ϕ ∼ 5 MP at the time of interest, we obtain (for λ ∼ 10−14 ; see below)

l(ϕ) ∼ exp(πϕ2 /M2P ), (7.5.25)

whereupon
M2P
ϕ2 ≈ ln l , (7.5.26)
π
where l here is measured in centimeters.
Equation (7.5.26) tells us that density perturbations on a scale lH ∼ 1028 cm are
produced at ϕ = ϕH , where

ϕH ≈ 4.5 MP ≈ 5.5 · 1019 GeV , (7.5.27)

and those on a galactic scale lg ∼ 1022 cm come into existence at ϕ = ϕg , where

ϕg ≈ 4 MP ≈ 5 · 1019 GeV . (7.5.28)

δρ λ
Equations (7.5.23) and (7.5.26) yield a general equation for in the theory ϕ4 :
ρ 4
√
δρ 2 6√
∼ λ ln3/2 l (cm) (7.5.29)
ρ 5π
with the amplitude of inhomogeneities on the scale of the horizon being
δρ √
∼ 150 λ , (7.5.30)
ρ
 132

while for those on a galactic scale,

δρ √
∼ 110 λ . (7.5.31)
ρ
δρ
The spectrum of is evidently almost flat, increasing slightly (logarithmically) at long
ρ
wavelengths.
We now discuss in somewhat more detail what the magnitude of the constant λ should
be in order for the predicted inhomogeneities to be consistent with the observational data
and the theory of galaxy formation.
Apparently, the most exacting constraints imposed by the cosmological data are not
δρ
on itself, but on the quantity A that determines the anisotropy of the microwave
ρ
∆T
background produced by adiabatic perturbations of the metric [223–227],
T
∆T A Kl
 
=q √ , (7.5.32)
T l l (l + 1) 10 π

∆T
where l is the order of the harmonic in the multipole expansion of (l ≥ 2 in (7.5.32)).
T
The relationship between A in (7.5.32) and metric perturbations is [220–222]
√
δρ(k) 2
= −2 Φ(k) = − α A(k) . (7.5.33)
ρ π

The numerical factors α and Kl in (7.5.32) and (7.5.33) depend on the specific assumptions
made about the nature of the missing mass in the universe. The magnitude of Kl is usually
of the order of one. As for α, that quantity is 2/3 for a hot universe, and 3/5 for a cold
one. In either case, 
π V3/2 
r
A(k) ≈ 16 π . (7.5.34)
3 3 dV k∼H(ϕ)

MP
dϕ
λ 4
In particular, for theϕ theory,
4
√
2 λ 3/2 √ √
A= ln l (cm) ∼ 1.2 λ ln3/2 l (cm) ∼ 6 · 102 λ (7.5.35)
3
∆T
on the scale of the horizon. From the observational constraints on , it follows that A
T
lies somewhere in the range

5 · 10−5 <
∼A<
−4
∼ 5 · 10 , (7.5.36)
 133

depending on the physical nature of the dark matter in the universe [227]. Condition
(7.5.2) is thus a consequence of (7.5.33) and (7.5.36) as well. To determine the constraints
on λ, it is most convenient to use (7.5.35) and (7.5.36) directly:

0.5 · 10−14 <

∼λ<
∼ 0.5 · 10
−12
. (7.5.37)

From here on, we assume for definiteness that

λ ∼ 10−14 , (7.5.38)

which is closer to estimates in the context of the theory of galaxy formation in a universe
filled with cold dark matter. As the theory of large scale structure in the universe is
∆T
developed and the observational limits on are refined [228], this estimate will improve.
T
Let us now consider another important example, the theory of a massive scalar field
m2 2
with V(ϕ) = ϕ . For a cold Friedmann universe in this case,
2
2 
δρ(k) 24 π m ϕ
r 

=  . (7.5.39)
ρ 5 3 MP MP 
k∼H

Likewise, for either a hot or cold universe,

√ r
π m

ϕ
2 

A(k) = 4 2π  . (7.5.40)
3 MP MP 
k∼H

√ λ 4
In this theory, both ϕH and ϕg are a factor of 2 less than in the ϕ theory. The analog
4
of Eq. (7.5.29) for the present theory is
δρ m
∼ 0.8 ln l [cm] , (7.5.41)
ρ MP
and on the scale of the horizon, Eq. (7.5.35) for A becomes
m
A ∼ 200 , (7.5.42)
MP
whence

3 · 1012 GeV ≈ 2.5 · 10−7 MP <

∼m<
−6 13
∼ 2.5 · 10 MP ≈ 3 · 10 GeV . (7.5.43)

Next, consider the more general theory with potential

λ ϕ4
n−4
ϕ

V(ϕ) = . (7.5.44)
n MP
For such a theory,
!1/2
π V(ϕ) ϕ
r
A = 16 π (7.5.45)
3 M4P n MP
 134

and the field ϕH is √

ϕH ∼ 2 n MP . (7.5.46)
Perturbations on the scale of the horizon are thus characterized by
!1/2
π V(ϕ)
r
A ∼ 32 π . (7.5.47)
3n M4P

Specifically, when A ∼ 10−4 , we find from (7.5.46) that in the last stages of inflation,
when the structure of the observable part of the universe had been formed, the value of
the effective potential was of order

V(ϕH ) ∼ 10−12 n MP ∼ n · 1082 g · cm−3 . (7.5.48)

The rate of expansion of the universe was then

√ √
H(ϕH ) ∼ 3 · 10−6 n MP ∼ 3.5 n · 1013 GeV ; (7.5.49)

that is, the universe increased in size by a factor of e in a time

t ∼ H−1 ∼ n−1/2 · 10−37 sec . (7.5.50)

In such a theory, the constant λ should be (for A ∼ 5 · 10−5 )

λ ∼ 2.5 · 10−13 n2 (4 n)−n/2 . (7.5.51)

These results give a general impression of the orders of magnitude which might be
encountered in realistic versions of the inflationary universe scenario. The estimate of
V(ϕH ) deserves special attention: a similar estimate can also be obtained from an analysis
of the theory of gravitational wave production at the time of inflation [117]. In the new
inflationary universe scenario, an analogous result implies that at all stages of inflation,
V(ϕ) should be at least ten to twelve orders of magnitude less than M4P [107, 229–231].
Within the framework of the chaotic inflation scenario, a similar statement is incorrect.
The value of A, which is 10−4 when ϕ ∼ ϕH , tends to increase in accordance with (7.5.45)
at large ϕ, and the observational data impose no upper limits whatever on V(ϕ). On the
other hand, we can derive a rather general constraint on the magnitude of V(ϕ) in the
last stages of inflation from (7.5.34). In fact, at the end of inflation, the rate at which the
potential energy V(ϕ) decreases becomes large — the energy density V(ϕ) is reduced by
a quantity that is O(V(ϕ)) within a typical time ∆t = H−1 . In other words, the criterion
Ḣ ≪ H2 is no longer satisfied. One can readily show that this means that at the end
V √
of inflation, V′ ∼ 8 π. In that case, (7.5.34) tells us that the quantity A, which
MP
is related to fluctuations of the field ϕ that are generated during the very last stage of
inflation, is given to order of magnitude by
v
u V(ϕ)
u
A∼ 10 t . (7.5.52)
M4P
 135

With A < −4
∼ 10 , we then find that at the end of inflation,
−10 4
V<
∼ 10 MP . (7.5.53)

This constraint applies both to the new inflationary universe scenario and the chaotic
inflation scenario.
The formalism that we have employed in this chapter rests on an assumption of the
δρ
relative smallness of . During the inflationary stage, as a rule, this condition is met.
ρ
λ
For example, in the ϕ4 theory,
4
√
δρ V′ δϕ 4 δϕ 2 H(ϕ) λϕ
∼ ∼ ∼ ∼ ≪1 (7.5.54)
ρ V ϕ πϕ MP
4
for V(ϕ) <∼ MP , λ ≪ 1.
On relatively small scales (l ∼ H−1 ), gradient terms ∂i (δϕ) ∂ i (δϕ) ∼ H4 make a sizable
δρ
contribution to . We have not considered these terms, since in the last analysis we were
ρ
interested in perturbations with exponentially long wavelengths. This contribution is also
much less than V(ϕ) when V(ϕ) ≪ M4P .
However, density perturbations produced at large ϕ become large after inflation. In
δρ λ
particular, according to (7.5.23), ∼ 1 in the ϕ4 theory for perturbations produced
ρ 4
when ϕ = ϕ∗ , where
ϕ∗ ∼ λ−1/6 MP . (7.5.55)
By (7.5.25), this means that after inflation ends, the universe only looks like a homoge-
neous Friedmann space on a scale
4
l∗ <
∼ exp(π λ
−1/3
) cm ∼ 106·10 cm . (7.5.56)

for λ ∼ 10−14 . This is many orders of magnitude larger than the observable part of the
universe, with lH ∼ 1028 cm, so for a present-day observer such inhomogeneities would lie
beyond his radius of visibility. From the standpoint of the global structure of the universe,
however, nonuniformity on scales l ≫ l∗ is exceedingly important, as we have discussed
in Chapter 1. We shall return to this question in Chapter 10.
We make one more remark in closing. We have been accustomed to calling the quan-
tity lH ∼ 3 t ∼ 1028 cm the distance to the horizon, as in the usual Friedmann model (see
(1.4.11)). But strictly speaking, the distance to the actual particle horizon in the infla-
tionary universe is exponentially large. Denoting this distance by RH (so as to distinguish
λ
it from lH ∼ 3 t), we may use (1.4.10) and (1.7.28) for the ϕ4 theory to obtain
4
π 107
RH ∼ M−1
P exp √ ∼ 10 cm (7.5.57)
λ
 136

(see also (1.7.39)). Nevertheless, this quantity can only tentatively be called the horizon.
The photons which presently enable us to view the universe only permit us to see back to
5
t>∼ 10 years after the end of inflation in our part of the universe, the reason being that
5
the hot plasma that filled the universe at t < ∼ 10 years was opaque to photons. Thus,
the size of that part of the universe accessible to electromagnetic observations is in fact
lH to high accuracy. A similar argument holds for neutrino astrophysics as well. We can
proceed a bit further, studying metric perturbations [136]. According to the standard
hot universe theory, gravitational waves provide the opportunity to obtain information
about any process in the universe that takes place at less than the Planck density, since
the universe is transparent to gravitational waves when T < ∼ MP . This is not true in the
inflationary universe scenario, however.
Let us consider a gravitational wave with wavelength l < ∼ lH (since these are the only
waves we can study experimentally). At the stage of inflation, when the scalar field ϕ
was equal to ϕH , the wavelength of this gravitational wave would have been of order
l ∼ H−1 ∼ 105 M−1 >
P (7.5.48), whereas at ϕ ∼ 1.05 ϕH , its wavelength would have been
−1
less than MP . Quantum fluctuations of the metric on this length scale are so large that
no measurements of gravitational waves inside the present horizon (at l ≤ lH ) could give
us any information about the structure of the universe with ϕ > ∼ 1.05 ϕH . In that sense,
MP 5
the range ϕ > ∼ 1.05 ϕH , corresponding to scales l >
∼ lH · H ∼ 10 lH , is “opaque” even to
gravitational waves. Thus, by analyzing perturbations of the metric, we can in principle
study phenomena beyond the visibility horizon (at l > lH ), but here we cannot progress
MP
beyond a factor of ∼ 105 . The energy density at the corresponding epoch (with
H(ϕH )
ϕ ∼ ϕH ) was seven orders of magnitude less than the Planck density (7.5.48). What
this means is that we cannot obtain information about the initial stages of inflation (with
V(ϕ) ∼ M4P ) — that is, the present state of the observable part of the universe is essentially
independent of the choice of initial conditions in the inflationary universe.

7.6 Are scale-free adiabatic perturbations sufficient

to produce the observed large scale structure
of the universe?
The creation of the theory of adiabatic perturbations in inflationary cosmology has been
an unqualified success. Beginning in 1982, when the theory was constructed in broad out-
line, theoretical investigations of the formation of large scale structure in the inflationary
universe have, as a rule, been based on two assumptions:
ρ
1) to high accuracy, the parameter Ω = is presently equal to unity (the universe is
ρ0
almost flat);
2) initial density perturbations leading to galaxy formation were adiabatic perturba-
δρ
tions with a flat (or almost flat) spectrum, ∼ 10−5 .
ρ
 137

The possibility of describing all of the existing data on large scale structure of the
universe on the basis of these simple assumptions is quite attractive, but we should recall
at this point the analogy between the universe and a giant accelerator. Experience has
taught us that the correct description of a large body of diverse experimental data is
seldom provided by the simplest possible theory. For example, the simplest description of
the weak and electromagnetic interactions would be given by the Georgi–Glashow model
[232], which is based on the symmetry group O(3). But the experimental discovery of
neutral currents forced us to turn to the far more complicated Glashow–Weinberg–Salam
model [1], based on the symmetry group SU(2) × U(1). The latter contains about 20
different parameters whose values are not grounded in any esthetic considerations at all.
For instance, almost all coupling constants in this theory are O(10−1 ), while the coupling
constant for interaction between the electron and the scalar (Higgs) field is 2 · 10−6. The
reason for the appearance of such a small coupling constant (just like the reason for the
appearance of the constant λ ∼ 10−14 in the simplest versions of the inflationary universe
scenario) is as yet unclear.
It seems unlikely that cosmology will turn out to be a much simpler science than
elementary particle theory. After all, the number of different types of large-scale objects
in the universe (quasars, galaxies, clusters of galaxies, filaments and voids, etc.) is very
large. The sizes of these objects form a hierarchy of scales that is absent from the flat
spectrum of the initial perturbations. In principle, some of these scales may be related
to the properties of the dark matter comprising most of the mass of the universe; see, for
example, [224, 235, 236]. Nevertheless, it is not at all obvious how to consistently describe
the formation of a large number of diverse large-scale objects, starting with the simple
assumptions (1) and (2). The requisite theory encounters a number of difficulties [235]
which, while not insurmountable, have nonetheless stimulated a search for alternative
versions of the theory of formation of large scale structure; see, for example, [236].
Another potential problem that a theory based on assumptions (1) and (2) may en-
∆T(θ)
counter is related to measurements of the anisotropy of the microwave background
T
∆T
radiation, where θ is the angular scale of observation. So far, only a dipole anisotropy
T
associated with the earth’s motion through the microwave background has been detected,
∆T
and neither a quadrupole anisotropy nor a small-angle anisotropy in has been found
T
∆T
at a level >
∼ 2 · 10−5 [228]. Meanwhile, flat-spectrum adiabatic density perturbations
T
∆T
should lead to > C · 10−5 [223–227], where the function C(θ) = O(1) depends on the
T ∼
angle θ and on the properties of dark matter. The function C(θ) is especially large at
∆T
large angles θ. The comparison between experimental constraints on and theoretical
T
predictions of quadrupole anisotropy is therefore a particularly important question, with a
δρ
bearing on perturbations on a scale l ∼ lH ∼ 1028 cm. The complexity of the situation
ρ
is exacerbated by the fact that the inflationary universe scenario gives an adiabatic per-
 138

δρ
turbation spectrum that is not perfectly flat. In most models, grows with increasing
ρ
λ
l. For example, in a theory with V(ϕ) ∼ ϕ4 , as we progress from the galactic scale lg to
4
δρ
the size of the horizon lH , the quantity increases by a factor of about 1.4 (see (7.5.29)
ρ
∆T
and (7.5.30)), resulting in a concomitant enhancement of the quadrupole anisotropy .
T
∆T
An assessment of the predictions of anisotropy in the simplest versions of the infla-
T
tionary universe scenario has already made it possible to discard the simplest models of
baryonic dark matter, and they cast some doubt upon the validity of models in which the
missing mass is concentrated in massive neutrinos [225, 226]. However, in the cold dark
matter models, in which the dark matter consists of axion fields [233, 234], Polonyi fields
[46, 15], or any weakly interacting nonrelativistic particles, the theoretical estimates of
∆T
are perfectly consistent with current observational limits [225, 226].
T
Thus, it remains possible that a theory of the formation of the large scale structure
of the universe can be completely assembled within the framework of the very simple
assumptions (1) and (2) (that is, a flat universe with a flat spectrum of adiabatic per-
turbations). However, as the inhabitants of new housing projects know only too well,
the simplest project is almost never the most successful. It would therefore be well to
understand whether we can somehow modify assumptions (1) and (2) while remaining
within the framework of the inflationary universe scenario. Specifically, we would like to
single out five basic questions.
1) Is it possible to get away from the condition Ω = 1?
2) Is it possible to obtain nonadiabatic perturbations after inflation?
3) Is it possible to obtain perturbations with a spectrum that decreases at l ∼ lH , so
∆T
as to reduce the quadrupole anisotropy of ?
T
4) Is it possible to obtain perturbations with a spectrum having one or perhaps sev-
eral maxima, which would help to explain the origin of the hierarchy of scales (galaxies,
clusters, . . . )?
5) Is it possible to produce the large scale structure of the universe through nonper-
turbative effects associated with inflation?
For the time being, the answer to the first question is negative: we know of no way to
obtain Ω 6= 1 in a natural manner within the context of inflationary cosmology. Even if
we could, it would most likely be only for some special choice of the potential V(ϕ) and
after painstaking adjustment of the parameters, for which there is as yet no particular
justification.
Building a model in which the spectrum of adiabatic perturbations falls off monotoni-
cally at long wavelengths is possible in principle, but rather difficult. The only reasonable
theory of this type that we are aware of is the Shafi–Wetterich model, based on a study of
inflation in the Kaluza–Klein theory [237]. A peculiar feature of this model is that infla-
 139

tion and the evolution of a scalar field ϕ (the role of which is played by the logarithm of
the compactification radius) are described by two different effective potentials, V(ϕ) and
W(ϕ). Unfortunately, it is difficult to realize the initial conditions required for inflation
in this model — see Chapter 9. Another suggestion that has been made is to study the
spectra produced by double inflation, driven first by one scalar field ϕ, then by another
Φ [238]. For the most natural initial conditions, however, the last stages of inflation are
governed by the field with the flattest potential (the smallest parameters m2 and λ). As
a rule, therefore, rather than leading to cutoff, two-stage inflation will lead to a more
δρ
abrupt rise in at the long wavelengths generated during the stage when the “heavier”
ρ
field ϕ is dominant.
Nevertheless, all of the questions posed above (except the first) can be answered in
the affirmative. There is a rather broad class of models which, in addition to adiabatic
perturbations, can also produce isothermal perturbations [239, 240], with spectra that
fall off at long wavelengths [239, 241]. Particularly interesting effects are associated with
phase transitions, which can occur at the later stages of inflation (when the universe still
has another factor of e50 –e60 left to expand). In particular, such phase transitions can
result in density perturbations having a spectrum with one or several maxima [242], and
to the appearance of exponentially large strings, domain walls, bubbles, and other objects
that can play a significant role in the formation of the large scale structure of the universe
[125, 243]. We shall discuss some of the possibilities mentioned above in the next two
sections.

7.7 Isothermal perturbations and adiabatic perturbations

with a nonflat spectrum
The theory of the formation of density perturbations discussed in Section 7.5 was based
upon a study of the simplest models, describing only a single scalar field ϕ responsible
for the dynamics of inflation. In realistic elementary particle theories, there exist many
scalar fields Φi of various kinds. To understand how inflation comes into play and what
sorts of density inhomogeneities arise in such theories, let us consider first the simplest
model, describing two noninteracting fields ϕ and Φ [239]:

1 1 m2 m2 λϕ 4 λΦ 4
L= (∂µ ϕ)2 + (∂µ Φ)2 − ϕ ϕ2 − Φ Φ2 − ϕ − Φ . (7.7.1)
2 2 2 2 4 4
We assume for simplicity that λϕ ≪ λΦ ≪ 1, and m2ϕ , m2Φ ≪ λϕ M2P . Then for large ϕ
and Φ, terms quadratic in the fields can be neglected. The only constraint on the initial
amplitudes of the fields ϕ and Φ is
λϕ 4 λΦ 4 4
V(ϕ) + V(Φ) ≈ ϕ + Φ <
∼ MP . (7.7.2)
4 4
 140

This means that the most natural initial values of the fields ϕ and Φ are ϕ ∼ λϕ−1/4 MP ,
−1/4
Φ ∼ λΦ MP ; that is, initially, V(ϕ) ∼ V(Φ) ∼ M4P , ϕ ≫ Φ ≫ MP . Since the curvature
of the potential V(Φ) is much greater than that of V(ϕ), it is clear that under the most
natural initial conditions, the field Φ and its energy V(Φ) fall off much more rapidly than
the field ϕ and its energy V(ϕ). The total energy density therefore quickly becomes equal
to V(ϕ); i.e., the Hubble parameter H(ϕ, Φ) becomes
s
2 π λϕ ϕ2
H(ϕ, Φ) ≈ H(ϕ) = . (7.7.3)
3 MP
Thus, within a short time, inflation will be governed solely by the field ϕ having the
potential V(ϕ) with the least curvature (the smallest coupling constant λϕ ). For this
reason, the field ϕ can be called the inflaton. It evolves just as if the field Φ did not exist
(see (1.7.21)):  s 
λϕ
ϕ(t) = ϕ0 exp − MP t . (7.7.4)
6π
In that case, the equation
Φ̈ + 3 H ϕ̇ = −λΦ Φ3 (7.7.5)
implies that during the inflationary stage
s
λϕ
Φ(t) = ϕ(t) , (7.7.6)
λΦ
and therefore q
3 MP 3 λϕ
m2 (ϕ) = m2 (Φ) = √ H(ϕ) , (7.7.7)
2π
where (for m2Φ ≪ λϕ M2P )
d2 V
m2 (ϕ) = = 3 λϕ ϕ2 ,
δϕ2
d2 V
m2 (Φ) = = 3 λΦ Φ2 . (7.7.8)
dΦ2
s
λϕ
In the last stage of inflation, ϕ ∼ MP and Φ ∼ MP . The perturbations of the ϕ and
λΦ
Φ fields have equal amplitudes (see (7.5.6)):
s
H λϕ ϕ2 q
δϕ = δΦ = = ∼ λϕ MP . (7.7.9)
2π 6 π MP
At that time, however, the contribution δρΦ that the field Φ makes to density inhomo-
geneities δρ is much less than δρΦ , the corresponding contribution from ϕ:
s s
dV λϕ λϕ
δρΦ = δΦ = λϕ ϕ3 δϕ = δρΦ ≪ δρϕ . (7.7.10)
dΦ λΦ λΦ
 141

It is therefore precisely the fluctuations δρϕ of the inflaton field that govern the am-
plitude of adiabatic density perturbations. At the stage of inflation with ϕ ∼ MP ,
δρϕ δρ 1 dV δϕ q
≈ = δϕ ∼ 4 ∼ λϕ , (7.7.11)
ρϕ ρ V δϕ ϕ
where ρ = ρϕ + ρΦ ≈ ρϕ . Meanwhile,
δρΦ 4 δΦ q
∼ ∼ λΦ . (7.7.12)
ρΦ Φ
After inflation, the inhomogeneities (7.7.11) produce the adiabatic perturbations (7.5.29),
which will in fact remain the dominant density perturbations if upon further expansion
of the universe, the quantity ρΦ falls off in the same way as ρϕ . However, in some cases
this condition is not satisfied, since the evolution of ρΦ and ρϕ depends on the interaction
of these fields with other fields, and on the shape of V(ϕ) and V(Φ). Let us assume, for
example, that the field Φ interacts very weakly with other fields. Such weakly interacting
scalar fields do exist in many realistic theories (axions, Polonyi fields, etc.). If the field ϕ
interacts strongly with other fields, its energy is quickly transformed into heat, ρϕ → T4R ,
and falls off with the expansion of the universe as T4 ∼ a−4 . At the same time, the field
Φ, without decaying, oscillates in the neighborhood of the point Φ = 0 with frequency
κ0 = mΦ . As it does so, its energy falls off in the same way as the energy of nonrelativistic
particles, ρΦ ∼ a−3 (see Section 7.9) i.e., much more slowly that the energy of the decay
products from the field ϕ. In the later stages of evolution of the universe, the energy of
the field Φ can therefore become greater than the energy of decay products of the inflaton
field, ρ = ρϕ + ρΦ ≈ ρΦ .
This is just the effect that underlies the possibility, discussed in [49], that the axion
field θ may be responsible for the missing mass of the universe at the present epoch.
Prior to the stage at which the field Φ is dominant, both the mean density ρΦ and the
δρΦ q
quantity ρΦ + δρΦ fall off in the same way, ρΦ ∼ ρΦ + δρΦ ∼ a−3 ; the quantity ∼ λΦ
ρΦ
therefore remains constant. From the start, inhomogeneities δρΦ are in no way associated
with the temperature inhomogeneities δT of decay products of the field ϕ, and in that
sense they are isothermal. They might also be called isoinflaton inhomogeneities, as they
are independent of fluctuations of the inflaton field ϕ. Consequently, due to the increasing
fraction of the overall matter density ρ accounted for by ρΦ , isothermal perturbations δρΦ ,
√ 2
q
with λΦ > ∼ 10 λϕ , begin to dominate, generating adiabatic perturbations
δρ δρΦ q
≈ ∼ λΦ (7.7.13)
ρ ρΦ
in the process. Note that in (7.7.13), there is no enhancement factor O(102 ) associated
with the transition from the inflationary stage to the expansion with a ∼ t1/2 or a ∼ t2/3 .
Thus, even in the simplest theory of two noninteracting fields, the process by which
density perturbations are generated can unfold in a fairly complicated manner: in addition
 142

to adiabatic density perturbations, isothermal perturbations can also come about, and for
λϕ ≪ 10−14 , λΦ > −10
∼ 10 , the latter can dominate.
Even more interesting possibilities are revealed when we allow for interactions between
the fields ϕ and Φ. Consider, for example, a theory with the effective potential
m2ϕ 2 λϕ 4 m2Φ 2 λΦ 4 ν 2 2
V(ϕ, Φ) = ϕ + ϕ − Φ + Φ + ϕ Φ + V(0) . (7.7.14)
3 4 2 4 2
Let us suppose that 0 < λϕ ≪ ν ≪ λΦ and λϕ λΦ > ν 2 , and assume also that m2ϕ ≪ λϕ M2P
and m2Φ ≪ C ν M2P , where C = O(1). Just as in the theory (7.7.1), the most natural initial
values for ϕ and Φ satisfy the conditions ϕ ≫ MP , ϕ ≫ Φ. With ϕ ≫ MP , the minimum
of V(ϕ, Φ) is located at Φ = 0, and the effective mass of the field Φ at Φ = 0 is

∂ 2 V 

m2Φ (ϕ, 0) = 2
= ν ϕ2 − m2Φ ≈ ν ϕ2 . (7.7.15)
∂Φ Φ=0 

This is much greater than the mass of the field ϕ,

m2ϕ (ϕ, 0) = m2ϕ + 3 λϕ ϕ2 ≈ 3 λϕ ϕ2 ≪ ν ϕ2 . (7.7.16)

The field Φ is therefore rapidly dumped into the minimum of V(ϕ, Φ), and as in the theory
(7.7.1), inflation becomes driven by the field ϕ.
In the last stage of inflation, when the field ϕ becomes less than
√
ϕc = C MP , (7.7.17)

the minimum of V(ϕ, Φ) is located at

m2Φ − ν ϕ2 CM2P − ϕ2
Φ2 = =ν , (7.7.18)
λΦ λΦ
and the effective mass of the field Φ is then

m2Φ (ϕ, Φ) = 2 ν (CM2P − ϕ2 ) . (7.7.19)

Notice that both when ϕ ≫ ϕc and q ϕ ≪ ϕc , the effective mass of the field Φ is much
λϕ ϕ2
greater than the Hubble constant H ∼ . Long-wave fluctuations δΦ of the field Φ
MP
are therefore generated only in some neighborhood of the phase transition point at ϕ ∼ ϕc .
In studying the density perturbations produced in this model, it turns out to be important
that the amplitude of fluctuations δΦ displays a variety of temporal behaviors, depending
on what precisely the value of the field ϕ was when these fluctuations arose. Numerical
calculations [242] taking this fact into account have shown that for certain relationships
among the parameters of the theory (7.7.14), the spectrum of adiabatic perturbations
produced during inflation may ! have a reasonably narrow maximum that is slightly shifted
π ϕ2c
with respect to l ∼ exp . .
M2P
 143

It should be pointed out here that many different types of scalar fields figure into
realistic elementary particle theories. It would therefore be hard to doubt that phase
transitions should actually take place at the time of inflation, and in fact most likely
not one, but many. The only question is whether these phase transitions take place
fairly late, when the field ϕ is changing between ϕH (7.5.27) and ϕg (7.5.28). This is
a condition that is satisfied, given an appropriate choice of parameters in the theory.
The actual parameter values chosen (like the parameters used in building the theories of
the weak and electromagnetic interactions) should be based on experimental data, rather
than on some a priori judgment about their naturalness (since according to that criterion
one could reject the Glashow–Weinberg–Salam model — see the preceding section). In
the case at hand, these data consist of observations of the large scale structure of the
universe and the anisotropy of the cosmic microwave radiation background. In our opinion,
the possibility of studying the phase structure of unified theories of elementary particles
and determining the parameters of these theories through astrophysical observations is
extremely interesting.
To conclude this section, let us briefly deal with the production of isothermal pertur-
bations in axion models. To this end, we examine the theory of a complex scalar field Φ
which interacts with an inflaton field ϕ:
m2ϕ 2 λϕ 4
V(ϕ, Φ) = ϕ + ϕ − m2Φ Φ∗ Φ
2 4
ν
+ λΦ (Φ∗ Φ)2 + ϕ2 Φ∗ Φ + V(0) . (7.7.20)
2
mΦ
After the spontaneous symmetry breaking which occurs at ϕ < ϕc = √ , the field Φ can
ν
be represented in the form !
i θ(x)
Φ(x) = Φ0 exp √ , (7.7.21)
2 Φ0
mΦ
where Φ0 = √ for ϕ ≪ ϕc . The field θ(x) is a massless Goldstone scalar field [244]
λΦ
with vanishing effective potential, V(θ) = 0.
In contrast to the familiar Goldstone field described above, the axion field is not
massless. Because of nonperturbative corrections to V(ϕ, Φ) associated with strong inter-
actions, the effective potential V(θ) becomes [233, 234]
!
Nθ
V(θ) = C m4π 1 − cos √ . (7.7.22)
2 Φ0
Here C ∼ O(1), and N is an integer that depends on the detailed structure of the theory;
for simplicity, we henceforth consider only the case N = 1. We thus see from (7.7.22) that
m2
axions can now have a small mass mθ ∼ π ∼ 10−2 GeV2 /Φ0 .
Φ0
From the standpoint of elementary particle theory, the main reason for considering the
axion field θ is that the field value that minimizes V(θ) automatically leads to cancellation
 144

of the effects of the strong CP violation that are associated with the nontrivial vacuum
structure in the theory of strong interactions [233, 234]. Cosmologists became interested
in this field for another reason. It turns out that at temperatures T ≫ 102 MeV, the
nonperturbative effects resulting in nonvanishing V(θ) are strongly √ inhibited. Therefore,
√
the field θ is equally likely to take any initial value in the range − 2 π Φ0 ≤ θ ≤ 2 π Φ0 .
2
As the temperature of the universe drops to T < ∼ 10 MeV, the effective potential V(θ)
takes the form (7.7.22), so that in the mean, the field θ acquires an energy density of
order m4π ∼ 10−4 GeV4 . This field interacts with other fields extremely weakly, and its
mass is extraordinarily small (mθ ∼ 10−5 eV for the realistic value Φ0 ∼ 1012 GeV; see
below). It therefore mainly loses its energy not through radiation, but through damping
of its oscillations near θ = 0 as the universe expands (by virtue of the term 3 H θ̇ in the
equation for the field θ). As we have already said, the energy density of any noninteracting
massive field, which oscillates near the minimum of its effective potential, falls off in the
same way as the energy density of a gas of nonrelativistic particles, ρθ ∼ a−3 , i.e., more
slowly than that of a relativistic gas. As a result, the relative contribution of the axion
field to the total energy density increases.
ρθ
The present-day value of the ratio depends on the value of Φ0 . For Φ0 ∼ 1012
ρ
GeV, most of the total energy density of the universe should presently be concentrated
in an almost homogeneous, oscillating axion field, which would then account for the
missing mass. As has been claimed in [49], a value of Φ0 ≫ 1012 GeV would be difficult
to reconcile with the available cosmological data (see Section 10.5, however). When
Φ0 ≪ 1012 GeV, the relative contribution of axions to the energy density of the universe
falls off as (Φ0 /1012 GeV)2 .
If the phase transition with symmetry breaking and the creation of the pseudo-
Goldstone field θ takes place during the inflationary stage, then inflation leads to fluctu-
H
ations of the field θ; as before, δθ = per unit interval ∆ ln k. When T < 102 MeV,
2π
δρθ δV(θ)
density inhomogeneities ∼ appear, which are associated with these fluctu-
ρθ V(θ)
ations. But because of the periodicity of the potential V(θ), these inhomogeneities are
related in a much more complicated way to the magnitude of the fluctuations in the field θ.
Let us assume, for example, that after a phase transition, inflation continues long enough
q H √
that the rms value hθ2 i = H t becomes much greater than Φ0 . The classical field
2π √ √
θ will then take on any value in the range − 2 π Φ0 ≤ θ ≤ 2 π Φ0 with uniform proba-
2
bility. When √ temperature drops down to T < 10 MeV, the surfaces at which the field θ
is equal to 2 Φ0 (2 n + 1) π form domain walls corresponding to the maximum of energy
density (7.7.22) [362]. Therefore, in addition to small isothermal density perturbations
proportional to perturbations δθ, in the axion cosmology one may get also exponentially
large (or even infinitely large) domain walls. In order to avoid disastrous cosmological
consequences of existence of such walls one should consider such theories where the Hub-
ble parameter H at the end of inflation becomes much smaller than Φ0 . In such a case the
probability of formation of domain walls becomes exponentially suppressed and no such
 145

walls appear in the observable part of the universe. It would be especially interesting to
investigate an intermediate case, where H is smaller than Φ0 , but not too much smaller.
In such a case the axionic domain walls will form small bubbles displaced exponentially
far away from each other. Is it possible that explosions of such bubbles may be responsible
for formation of the large-scale structure of the universe? We will return to a discussion
of similar possibilities in the next section.
Another interesting possibility which may appear in the axion cosmology is related to
the time-dependence of the radius of the axion potential Φ0 . Indeed, isothermal pertur-
bations generated in the axion cosmology are proportional to perturbations of the angle,
δρθ δθ
∼ √ . During inflation, the value of Φ0 in the model (7.7.20) rapidly changes.
ρθ 2 Φ0
This leads to the modification of the spectrum of density perturbations, which acquires
a strong maximum on the scale corresponding to the moment of the phase transition (at
mΦ
which ϕ = ϕc = √ , Φ0 = 0). Again, one should worry about overproducing density per-
V
turbations and axionic domain walls on this scale [362]. Our main purpose in discussing
all these possibilities was just to demonstrate that by a simple extension of minimal infla-
tionary models containing only one scalar field one can easily obtain perturbations with
a non-flat spectrum, either adiabatic or isothermal. One should not use such possibilities
without a demonstrated need, but it is better to know that if the experts in the theory
of galaxy formation will tell us that they are unhappy with adiabatic perturbation with
a flat spectrum, inflation will have something else to offer.

7.8 Nonperturbative effects: strings, hedgehogs, walls,

bubbles, . . .
In Section 7.5 we studied mechanisms for generating small density perturbations in the
inflationary universe. However, as we have seen in the previous section, phase transitions
at the time of inflation can lead not just to small density perturbations; they can also
produce nontrivial structures of exponentially large size. Herewith, we present some
examples.
1. Strings. The theory of the formation of density inhomogeneities during the evolution
of cosmic strings [81] was long considered to be the only real alternative to inflationary
theory for the formation of flat-spectrum adiabatic perturbations. It is now quite clear
that there exists a wide range of other possibilities — see Section 7.7 and the discussion
below. Furthermore, without inflation, string theory provides no help in solving the
problems of standard Friedmann cosmology, and the formation of superheavy strings
through high-temperature phase transitions following inflation is complicated by the fact
that in most models, the universe after inflation is not hot enough. But it is perfectly
possible to produce strings during phase transitions in the inflationary stage [125, 246,
247]. About the simplest model in which one could treat such a process would be a theory
describing the interaction of an inflaton ϕ with a complex scalar field Φ having effective
 146

m2Φ
potential (7.7.20). In the early stages of inflation, when ϕ2 > , symmetry is restored in
ν
mΦ
the theory (7.7.20). As the field ϕ falls off toward ϕ = ϕc = √ , the symmetry-breaking
ν
phase transition leading to the production of strings takes place, as happens in the case
of a phase transition with a decrease in temperature; see Section 6.2. The difference
here is that during inflation, the
! typical size of the strings produced increases by a factor
π ϕ2c π m2Φ
!
of exp = exp . If this factor is not too large, then most of the results
M2P ν M2P
obtained in the theory of formation of density inhomogeneities due to strings [81] remain
valid.
2. Hedgehogs. Phase transitions at the time of inflation also lead to the production
of hedgehog-antihedgehog pairs (see Section 6.2). A typical separation r0 between a
hedgehog and antihedgehog is of order H−1 (ϕc ), but as a result of inflation, this separation
increases exponentially. The pair energy is proportional to r. Hedgehog annihilation
begins when the size of the horizon ∼ t growth to a size comparable with the distance
δρ
between the hedgehog and antihedgehog. This gives rise to density inhomogeneities
ρ
Φ20
of order 2 , for the same reason as in the theory of strings (6.2.3). In the present case,
MP
however, the spectrum of density inhomogeneities will have a pronounced maximum at a
π ϕ2c
!
wavelength of the order of the typical distance between hedgehogs, ∼ exp .
M2P
3. Monopoles. Monopoles can be produced as a byproduct of phase transitions at
the time of inflation. The density of such monopoles will be reduced by factors like
3 π ϕ2c
!
exp − , but for sufficiently small ϕc , attempts to detect them experimentally
M2P
may have some chance of succeeding.
4. Monopoles connected by strings. Such objects also crop up in certain theories. Just
as for hedgehogs, such monopoles appear in a confinement phase, and in the hot universe
theory, where the typical distance between monopoles is of order T−1 c , they are rapidly
annihilated [81]. In the inflationary universe scenario, they can lead to approximately the
same consequences as hedgehogs.
5. Domain walls bounded by strings. The axion theory discussed in the preceding
section qualifies as one of a number of theories in which strings are produced following
symmetry breaking. With currently accepted model parameter values, axion strings, in
and of themselves, are too light to induce large enough density inhomogeneities. But a
more careful analysis shows that every axion string is the boundary of a domain wall [43,
θ(x)
81]. This is related to the fact that in proceeding around a string, with the quantity √
2 Φ0
changing by 2 π, we necessarily pass through a maximum of V(θ) (7.7.22). Energetically,
the most favorable configuration of the field θ(x) is that in which the field θ does not
change as one proceeds around the string; this corresponds to having a minimum of V(θ)
everywhere except at a wall whose thickness is of order m−1 θ , and having the quantity
 147

θ(x)
√ change by 2 π when the latter is traversed. The surface energy of the wall is of
2 Φ0
order m2π Φ0 .
Analysis of the evolution of a system of strings that acts like a wireframe supporting
a soap film composed of domain walls shows that the initial field configuration resembles
a single infinitely curved surface containing a large number of holes. Moreover, there also
exist isolated surfaces of finite size, but they contribute negligibly to the total energy of
the universe [81]. Portions of these surfaces eventually begin to intersect and tear one
another over a surface region of small extent, and resemble frothy “pancakes,” which
subsequently oscillate and radiate their energy away in the form of gravitational waves.
If these surfaces form as a result of phase transitions in a hot universe, the pancakes turn
out to be extremely small, and they quickly disappear. But surfaces created at the time
of inflation produce pancakes that are exponentially large [81, 125]. The possible role
of such objects in the formation of large scale structure in the universe requires further
investigation.
6. Bubbles. In studying the cosmological consequences of the phase transitions occur-
ring at the time of inflation, we have implicitly assumed that they were soft transitions,
with no tunneling through barriers, as in the second-order phase transition considered in
Section 7.7. Meanwhile, the phase transitions can also be first-order — see Section 7.4.
Bubbles of the field Φ could then be produced.
During inflation, there is much less energy in the fields Φ than in the inflaton field
ϕ. For that reason, the appearance of such bubbles has practically no effect on the rate
of expansion of the universe, and after inflation, the sizes of bubbles of the field Φ wind
up being exponentially!large; more specifically, all bubbles turn out to have a typical
π ϕ2c
size of order exp cm. If the rate of bubble production is high, then the resulting
M2P
distribution of the field Φ will resemble a foam (cells), with maximum energy density on
the walls of adjoining bubbles and with voids within them. If the bubble creation rate is
low, then mutually separated regions will arise within which the matter density is lower
than that outside. In the later stages of the evolution of the universe, when the energy
of the field Φ may become dominant, the corresponding density contrast may turn out to
be quite high [125, 240, 243].
7. Domains. Especially interesting effects can transpire when the universe acquires
a domain structure at the time of inflation. As the simplest example of this we consider
the possible kinetics of the SU(5)-breaking phase transition at that epoch. As we already
noted in the preceding chapter, as the temperature T drops, phase transitions in the SU(5)
theory entail the formation of bubbles which can contain a field Φ that corresponds to any
one of four different types of symmetry breaking: SU(3) × SU(2) × U(1), SU(4) × U(1),
SU(3) × U(1) × U(1), or SU(2) × SU(2) × U(1) × U(1). An analogous phase transition
can also take place during inflation, but in the latter event, inflation ensures that bubbles
of the different phases becomes exponentially large. This results in the formation of
large domains filled with matter in different phases — that is, with slightly different
density. In the standard SU(5) model, only the phase SU(3) × SU(2) × U(1) is stable after
 148

inflation, so ultimately the entire universe is transformed to this phase, and the domain
walls that separate the different phases disappear. However, the corresponding density
inhomogeneities that appeared during the epoch when domains were still present somehow
remain imprinted on the subsequent density distribution of matter in the universe.
If the probability of bubble formation is significantly different for bubbles containing
matter in different phases, the universe will eventually consist of islands of reduced or
enhanced density superimposed on a relatively uniform background. In principle, we could
associate such islands with galaxies, clusters of galaxies, or even the insular structure of
the universe proposed in [248].
If on the other hand the universe simultaneously spawns comparable numbers of bub-
bles in different phases, the resulting density distribution takes on a sponge-like structure.
Specifically, there will be cells containing phases of different density, but a significant frac-
tion of the cells of a given phase will be connected to one another, so that one could pass
from one part of the universe to another through cells that are all of the same type (per-
colation). Concepts involving a sponge-like universe have lately become rather popular.
Recent results [249] indicating that the universe effectively consists of contiguous bub-
bles 50–100 Mpc (1.5 · 1026 –3 · 1026 cm) in size containing few luminous entities, so that
galaxies are basically concentrated at bubble walls, have been especially noteworthy. Par-
ticularly interesting in that regard is the fact that such structures may appear as a natural
consequence of phase transitions at the time of inflation [125, 244].
In the context of the present model, the advent of regions of the universe containing
most of the luminous (baryon) matter is not at all necessarily connected with density
enhancements above the mean. Firstly, post-inflation baryon production (see the next
section) proceeds entirely differently in the different phases (SU(3) × SU(2) × U(1) or
SU(4) × U(1)). It could turn out, in principle, that baryons are only produced in those
regions filled with a phase whose density lies below the mean, and these would then be
just the regions in which we would see galaxies. Secondly, if galaxy formation is associated
with isothermal perturbations of the field Φ, then one should take into account that the
amplitude of such perturbations will also depend on the phase inside each of the domains.
Isothermal perturbations can therefore only be large enough for subsequent galaxy forma-
tion in those domains filled with some particular phase, and these are precisely the regions
in which galaxies, clusters of galaxies, and so forth should form. Thus, depending on the
specific elementary particle theory chosen, galaxies will preferentially form in regions of
either enhanced or reduced density, and either in the space outside bubbles (for example,
at bubble walls) or within them.
If certain phases remain metastable after inflation, then as a rule their characteristic
decay time will turn out to be much greater than the age of the observable part of the
universe, t ∼ 1010 yr. In that event, the universe should be partitioned right now into
domains that contain matter in a variety of phase states. This is just the situation in
supersymmetric SU(5) models, where the minima corresponding to SU(5), SU(3)×SU(2)×
U(1), and SU(4) × U(1) symmetries are of almost identical depth and are separated from
one another by high potential barriers [91–93]. During inflation, the universe is partitioned
into exponentially large domains, each of which contains one of the foregoing phases, and
 149

we happen to live in one such domain corresponding to the SU(3) × SU(2) × U(1) phase
[211]. If inflation were to go on long enough after the phase transition (that is, if the phase
λ 4
transition had taken place with ϕc > ∼ 5 MP in the 4 ϕ theory), then there would be not
a single domain wall in the observable part of the universe. In the opposite case, domains
would be less than 1028 cm in size. If regions containing different phases have the same
probability of forming (as in a theory that is symmetric with respect to the interchange
ϕ → −ϕ), then for ϕc < ∼ 5 MP, we encounter the domain wall problem discussed in Section
6.2. However, the probability of producing bubbles containing matter in different phases
depends on the height of the walls separating different local minima of V(Φ), and in
general differs significantly among the phases. The universe is therefore mostly filled with
just one of its possible phases, and the other phases are present in the form of widely-
spaced, exponentially large, isolated domains. Those domains containing energetically
unfavorable phases later collapse. As we have already remarked in Section 7.4, those
regions whose probability of formation is fairly low should be close to spherical, and the
collapse of such regions proceeds in an almost perfectly spherically symmetric manner.
The entire gain in potential energy due to compression of a bubble of a metastable phase
would then be converted into kinetic energy of its collapsing wall. If the wall is comprised
of a scalar field Φ that interacts strongly enough both with itself and with other fields, the
bubble wall energy will be transformed after collapse into the energy of those elementary
particles created at the instant of wall collapse. The particles thus produced fly off in all
directions, forming a spherical shell. We thereby have yet another mechanism capable of
producing a universe with bubble-like structure. The process will be more complicated if
the original bubble is significantly nonspherical, and the cloud of newly-created particles
will also no longer be spherically symmetric.
This model resembles the model of Ostriker and Cowie [250] for the explosive formation
of the large scale structure of the universe. However, the physical mechanism discussed
above differs considerably from that suggested in Ref. 250.
The investigation of nonperturbative mechanisms for the formation of the large scale
structure of the universe has just begun, but it is already apparent from the foregoing
discussion how many new possibilities the study of the cosmological consequences of phase
transitions during the inflationary stage carries with it. The overall conclusion is that in-
flation can result in the appearance of various exponentially large objects. The latter may
be of interest not just as the structural material out of which galaxies could subsequently
be built, but, for example, as a possible source of intense radio emission [251]. They could
also turn into supermassive black holes, and finally they might turn out to be responsible
for anomalous exoergic processes in the universe. This abundance of new possibilities
does not mean that anything goes, but it does substantially expand the horizons of those
seeking the correct theory of formation of the large scale structure of the universe.
 150

7.9 Reheating of the universe after inflation

The production of inhomogeneities in the inflationary universe has elicited a great deal of
interest of late, as this process is directly reflected in the structure of the observable part of
the universe. Of no less value is the study of the process whereby the universe is reheated
and its baryon asymmetry generated, since this process is a mandatory connecting link
between the inflationary universe in its vacuum-like state and the hot Friedmann universe.
In the present section, we study the reheating process through the example of the simplest
theory of a massive scalar field ϕ that interacts with a scalar field χ and a spinor field ψ,
with the Lagrangian

1 m2 1 m2
L = (∂µ ϕ)2 − ϕ ϕ2 + (∂µ χ)2 − χ χ2 + ψ̄ (i γµ ∂µ − mψ ) ψ
2 2 2 2
+ ν σ ϕ χ2 − h ψ̄ ψ ϕ − ∆V(ϕ, χ) . (7.9.1)

Here ν and h are small coupling constants, and σ is a parameter with the dimensionality
of mass. In realistic theories, the constant part of the field ϕ, for example, can play the
role of σ. What we mean by ∆V(ϕ, χ) is that part of V(ϕ, χ) that is of higher order in
ϕ2 and χ2 . We shall assume (allowing for ∆V(ϕ, χ)) that in the last stages of inflation,
the role of the inflaton field is taken on by the field ϕ, and then go on to investigate
the process by which the energy of this field is converted into particles χ and ψ. We
suppose for simplicity that mϕ ≫ mχ , mψ , and that at the epoch of interest, ν σ ϕ ≪ m2χ ,
h ϕ ≪ mψ .
If we ignore effects associated with particle creation, the field ϕ after inflation will
oscillate near the point ϕ = 0 at a frequency k0 = mϕ . The oscillation amplitude will
fall off as [a(t)]−3/2 , and the energy of the field ϕ will decrease in the same way as the
m2ϕ 2
density of nonrelativistic ϕ particles of mass mϕ : ρϕ = V(ϕ) = ϕ ∼ a−3 , where ϕ
2
is the amplitude of oscillations of the field [252]. The physical meaning of this is that a
homogeneous scalar field ϕ, oscillating at frequency mϕ , can be represented as a coherent
ρϕ mϕ 2
wave of ϕ-particles with vanishing momenta and particle density nϕ = = ϕ . If
mϕ 2
the total number of particles ∼ nϕ a3 is conserved (no pair production), the amplitude of
the field ϕ will fall off as a−3/2 . The equation of state of matter at that time is p = 0; i.e.,
2
a(t) ∼ t3/2 , H = , ϕ ∼ a−3/2 ∼ t−1 .
3t
In order to describe the particle production process with its concomitant decrease in
the amplitude of the field ϕ, let us consider the quantum corrections to the equation of
motion for the homogeneous field ϕ, oscillating at a frequency k0 = mϕ ≫ H(t):

ϕ̈ + 3 H(t) ϕ̇ + [m2ϕ + Π(k0 )] ϕ = 0 . (7.9.2)

Here Π(k0 ) is the polarization operator for the field ϕ at a

four-momentum k = (k0 , 0, 0, 0), k0 = mϕ .
The real part of Π(k0 ) gives only a small correction to m2ϕ , but when k0 > 2 mχ (or
k0 > 2 mψ ), Π(k0 ) acquires an imaginary part Π(k0 ). For m2ϕ ≫ H2 and m2ϕ ≫ Im Π, and
 151

neglecting the time-dependence of H, we obtain a solution of Eq. (7.9.2) that describes

the damped oscillations of the field ϕ near the point ϕ = 0:
" ! #
1 Im Π(mϕ )
ϕ = ϕ0 exp(i mϕ t) · exp − 3H+ t . (7.9.3)
2 mϕ
From the unitarity relations [10, 124], it follows that
Im Π(mϕ ) = mϕ Γtot , (7.9.4)
where Γtot is the total decay probability for a ϕ-particle. Hence, when Γtot ≫ 3 H,
the energy density of the field ϕ decreases exponentially in a time less than the typical
expansion time of the universe ∆t ∼ H−1 :
m2 ϕ2
ρϕ = ∼ ρ0 e−Γtot t . (7.9.5)
2
This is exactly the result one would expect on the basis of the interpretation of the
oscillating field ϕ as a coherent wave consisting of (decaying) ϕ-particles.
The probability of decay of a ϕ-particle into a pair of χ-particles or ψ-particles is
known — see, for example, [10, 122, 123]. For mϕ ≫ mχ , mψ ,
ν2 σ
Γ(ϕ → χ χ) = , (7.9.6)
8 π mϕ
h2 mϕ
Γ(ϕ → ψ̄ ψ) = . (7.9.7)
8π
If the constants ν σ and h2 are small, then initially Γtot = Γ(ϕ → χ χ) + Γ(ϕ → ψ̄ ψ) <
2
3 H(t) = . Basically, in that event, the energy density of the field ϕ decreases from the
t
m2 ϕ2
outset simply due to the expansion of the universe, ∼ t−2 . The fraction of the total
2
energy converted into energy of the particles that are produced remains small right up to
the time t∗ at which 3 H(t∗ ) becomes less than Γtot . Particles produced prior to this time
can also be thermalized, in principle, and their temperature in certain situations can be
even higher than the final temperature TR [253]. But the contribution of newly-created
particles to the overall matter density becomes significant only starting at the time t∗ ,
after which practically all the energy of the field ϕ is transformed into the energy of newly-
created χ- and ψ-particles within a time ∆t ∼ t∗ < −1 ∗
∼ H . The condition 3 H(t ) ∼ Γtot
∗
tells us that the energy density of these particles at the time t is of order
∗ Γ2tot M2P
ρ ∼ . (7.9.8)
24
If the χ- and ψ-particles interact strongly enough with each other, or if they can rapidly
decay into other species, then thermodynamic equilibrium quickly sets in, and matter
acquires a temperature TR , where according to (1.3.17) and (7.9.8)
π N(TR ) 4 Γ2 M2
ρ∗ ∼ TR ∼ tot P (7.9.9)
30 24
 152

Here N(TR ) is the effective number of degrees of freedom at T = TR , with N(TR ) ∼ 102 –
103 , so that q
TR ∼ 10−1 Γtot MP . (7.9.10)
Note that as we said earlier, TR does not depend on the initial value of the field ϕ, and
is determined solely by the parameters of the elementary particle theory.
δρ
Let us now estimate TR numerically. In order for adiabatic inhomogeneities ∼ 10−5
ρ
to appear in the present theory, it is necessary that mϕ be of order 10−6 MP ∼ 1013
GeV. One can readily verify that quantum corrections investigated in Chapter 2 do not
2 8 mϕ −5
significantly alter the form of V(ϕ) at ϕ <
∼ MP only if h <∼ M ∼ 10 and ν σ < ∼ 5 mϕ ∼
P
1014 GeV. Under these conditions,

Γ(ϕ → χ χ) <
∼ mϕ ∼ 10−6 MP , (7.9.11)
m2ϕ
Γ(ϕ → ψ̄ ψ) <
∼ ∼ 10−12 MP . (7.9.12)
MP
For completeness, we note that in theories like the Starobinsky model or supergravity,
the magnitude of Γ for ϕ-field decays induced by gravitational effects is usually [135, 286]

m3ϕ
Γg ∼ ∼ 10−18 MP . (7.9.13)
M2P

Thus, if direct decay of the field ϕ into scalar χ-particles is possible, then one might expect
that in general this will be the leading process [123]. We see from (7.9.11) that the rate
at which the ϕ-field decays into χ-particles can be of the same order of magnitude as the
rate at which the ϕ-field oscillates; it can therefore divest itself of most of its energy in
several cycles of oscillation (or even simply in the time it takes to roll down from ϕ ∼ MP
to ϕ = 0 [254]). Since H(ϕ) ∼ mϕ at the end of inflation, the universe has almost no
time to expand during reheating, and almost all the energy stored in the field ϕ can be
converted into energy for the production of χ-particles. This same result follows from
(7.9.8):
∗ m2ϕ 2 m2ϕ MP
ρ = ϕ ∼< , (7.9.14)
2 24
whence ϕ(t∗ ) <∼ MP and
q
TR <
∼ 10
−1
mϕ MP ∼ 1015 GeV . (7.9.15)

Reheating to TR ∼ 1015 GeV occurs only for a special choice of parameters. Moreover, in
some models the universe cannot be heated up to a temperature much higher than mϕ ;
see Section 7.10. Nevertheless, one must keep in mind the possibility of such an efficient
reheating, which can take place immediately after inflation ends, despite the weakness of
the interaction between the ϕ- and χ-fields. A similar possibility can come into play if
 153

the potential V(ϕ) takes on a more complicated form — for example, if the curvature of
V(ϕ) near its minimum is much greater than at ϕ ∼ MP [255].
If the field ϕ can only decay into fermions, then we find from (7.9.12) and (7.9.10)
that in the simplest models, the temperature of the universe after reheating will be at
least three orders of magnitude lower,
−1 12
TR <
∼ 10 mϕ ∼ 10 GeV , (7.9.16)

and if gravitational effects are dominant, then

s
mϕ
TR <
∼ 10
−1
mϕ ∼ 109 GeV . (7.9.17)
MP
The foregoing estimates have been made using the simplest model and assuming that
the oscillating field is small. But if the field ϕ is large (ν σ ϕ > m2χ or h ϕ > mψ ), it is
not enough to calculate just the polarization operator; one must then either calculate the
imaginary part of the effective action S(ϕ) in the external field ϕ(t) [122, 256], or employ
methods based on the Bogoliubov transformation [74].
We shall not discuss this point in detail here, since for the theory (7.9.1), an inves-
tigation of the case in which ν σ ϕ > m2χ and h ϕ > mψ results only in a change in the
numerical coefficients in (7.9.6) and (7.9.7). More important changes arise in the theories
with Lagrangians that lack ternary interactions like ϕ χ2 and ϕ ψ̄ ψ, having only vertices
like ϕ4 , ϕ2 χ2 or ϕ2 A2µ , with a field ϕ that has no classical part ϕ0 .
λ
Thus, for example, in the ϕ4 theory of the massless field, evaluation of the imaginary
4
part of the effective Lagrangian L(ϕ) leads to an expression for the probability of pair
production [122]:
P ≈ 2 Im L(ϕ) ∼ λ2 ϕ4 · O(10−3 ) . (7.9.18)
An analogous expression holds for a λ ϕ2 χ2 theory. The energy density of particles created
in a time ∆t ∼ H−1 is
√
∆ρ ∼ 10−3 λ2 ϕ4 · λ ϕ · H−1 ∼ 10−3 λ2 ϕ3 MP , (7.9.19)
√
where the effective mass of the ϕ- and χ-fields is O( λ ϕ). This quantity becomes com-
λ
parable to the total energy density ρ(ϕ) ∼ ϕ4 when
4
−2
ϕ<
∼ 10 λ MP , (7.9.20)

that is, when

ρ(ϕ) ∼ 10−8 λ5 M4P , (7.9.21)
whereupon
−3 5/4
TR <
∼ 10 λ MP . (7.9.22)
For λ ∼ 10−14 ,
−21 −2
TR <
∼ 3 · 10 MP ∼ 3 · 10 GeV . (7.9.23)
 154

λ 4
If the field ϕ in the theory ϕ has a nonvanishing mass mϕ , the reheating of the universe
4
mϕ
becomes ineffectual at ϕ < ∼ √λ , since at small ϕ, the value of ∆ρ from (7.9.19) always
m2ϕ 2
turns out to be less than ρ(ϕ) ∼ ϕ . In such a situation, the energy of the field ϕ
2
basically falls due to the expansion of the universe, ρ(ϕ) ∼ a−3 , rather than due to the
field decay. This implies that after expansion of the universe to its present state, even a
strongly interacting oscillating classical field ϕ (10−14 ≪ λ <
∼ 1) can turn out to be largely
undecayed into elementary particles, and can make a sizable contribution to the density
of dark matter in the universe. This topic is treated in more detail in [254].

7.10 The origin of the baryon asymmetry of the universe

As we have already noted, the elaboration of feasible mechanisms for generating an excess
of baryons over antibaryons in the inflationary universe [36–38] was one of the most im-
portant stages in the development of modern cosmology. The baryon asymmetry problem
served to demonstrate quite clearly that questions which to many had seemed meaningless,
or at best metaphysical (“Why is the universe structured as it is, and not otherwise?”),
could actually have a physical answer. Without a solution of the baryogenesis problem,
the inflationary universe scenario would be impossible, since the density of baryons that
exist at the earliest stages of evolution of the universe becomes exponentially small after
inflation. The generation of a baryon asymmetry in the universe is therefore just as indis-
pensable an element of the inflationary universe scenario as the reheating of the universe
discussed in the previous section.
As the first treatment of the origin of baryons in the universe made clear [36], an asym-
metry between the number of baryons and
antibaryons arises when three conditions are satisfied:
1. The processes involved violate baryon charge conservation.
2. These processes also violate C and CP invariance.
3. Baryon production processes take place in a nonequilibrium thermodynamic state.
One example would be the decay of particles with mass M ≫ T.
The need for the first condition is obvious. The second is needed in order for the decay
of particles and antiparticles to produce different numbers of baryons and antibaryons.
The third condition is primarily needed to prevent inverse processes which might destroy
baryon asymmetry.
Genuine interest in the possibility of generating the baryon asymmetry of the uni-
verse was kindled by the advent of grand unified theories, in which baryons could freely
transform into leptons prior to symmetry breaking between the strong and electroweak
interactions. After the symmetry breaking, superheavy scalar and vector particles (Φ,
H, X, and Y) decay into baryons and leptons. If the decay of these particles takes place
in a state far removed from thermodynamic equilibrium, so that the inverse processes
 155

of baryon and lepton reversion to superheavy particles are inhibited, and if C and CP
invariance is violated, then the decays will produce slightly different numbers of baryons
and antibaryons. This difference, after annihilation of baryons and antibaryons, is ex-
actly what produces the baryonic matter that we see in our universe. The small number
nB
∼ 10−9 comes about as a product of the gauge coupling constant in grand unified
nγ
theories, the constant related to the strength of CP violation, and the relative abundance
of the particles that produce baryon asymmetry after their decay [38].
We shall not dwell here on a detailed description of this mechanism for baryogenesis,
referring the reader instead to the excellent reviews in [105, 257, 258]. What is important
nB
for us is that theories leading to the desired result ∼ 10−9 actually do exist. A similar
nγ
mechanism can also operate as a part of the inflationary universe scenario, where it is
even more effective, since the universe is reheated after inflation in what is essentially a
nonequilibrium process, and during this process, superheavy particles can be produced
with masses much greater than the temperature of the universe after reheating TR [122].
However, in the minimal SU(5) theory with a single family of Higgs bosons H and the
nB
most natural relationship between coupling constants, turns out to be many orders of
nγ
nB
magnitude less than 10−9 . In order to get ∼ 10−9 , it is necessary either to introduce
nγ
two additional families of Higgs bosons, or to consider the possibility of a complex sequence
of phase transitions as the universe cools in the SU(5) theory [259]. Furthermore, it is far
from easy to obtain the fairly large number of superheavy bosons needed to implement
this mechanism at the time of post-inflation reheating. This is especially difficult in
supergravity-type theories, where the temperature to which the universe is reheated is
usually 1012 GeV at most. Finally, one more potential problem became apparent fairly
recently. It was found that nonperturbative effects lead to efficient annihilation of baryons
and leptons at a temperature that is higher than or of the same order as the phase
transition temperature Tc ∼ 200 GeV in the Glashow–Weinberg–Salam model [129].
This means that if equal numbers of baryons and leptons are generated in the early stages
of evolution of the universe, so that B − L = 0, where B and L are the baryon and
lepton charges respectively (and this is exactly the situation in the simplest models of
baryogenesis [38]), then the entire baryon asymmetry of the universe arising at T > 102
GeV subsequently vanishes. If this is so, then either it is necessary to have theories that
begin with an asymmetry B − L 6= 0, which makes these theories even more complicated,
or mechanisms for baryogenesis which could operate efficiently even at a temperature
2
T< ∼ 10 GeV must be worked out. Several possible mechanisms of this kind have been
proposed in recent years. Below we describe one of them, the details of which are probably
the closest of any to the inflationary universe scenario.
The basic idea behind that scheme was proposed in a paper by Affleck and Dine [97];
their mechanism was implemented in the context of the inflationary universe scenario
[98]. Later, it was demonstrated that this mechanism could work in models based on
superstring theory [260]. Referring the reader to the original literature for details, we
 156

discuss here the basic outlines of this new mechanism for baryogenesis.
As an example, consider a supersymmetric SU(5) grand unification theory. In this
theory there exist squarks and sleptons, which are scalar fields, the superpartners of quarks
and leptons. An analysis of the shape of the effective squark and slepton potential shows
that it has valleys — flat directions — in which the effective potential approaches zero
[97]. We will refer to the corresponding linear combinations of squark and slepton fields
in the flat directions as the scalar field ϕ. After supersymmetry breaking in the model in
question, the value of the effective potential V(ϕ) in the valleys rises slightly, and the field
ϕ acquires an effective mass m ∼ 102 GeV. Excitations of this field consist of electrically
neutral unstable particles having baryon and lepton charge B = L = ±1. The baryon
charge of each such particle is not conserved by their interactions, but the difference B−L
is. These particles interact among themselves via the same gauge coupling constant g as
do the quarks. The coupling constant of the baryon-nonconserving interactions of the ϕ-
m2
!
particles is λ = O , where MX is the X-boson mass in the SU(5) theory. For large
M2X
values of the classical field ϕ, many of the particles that interact with it acquire a very
high mass that is O(g ϕ). However, there are also light particles such as quarks, leptons,
W mesons, and so forth, that interact only indirectly with the field ϕ (via radiative
m2 g2
 2
αs
corrections), with an effective coupling constant λ̃ ∼ , where α s = . In a
π ϕ2 4π
rigorous treatment, it would be necessary to consider the dynamics of the two fields v
and a, corresponding to different combinations of squark-slepton fields in the valley of
the effective potential [97]. A thorough study of a system of such fields in the SU(5)
theory would be fairly complicated, but fortunately, in the most important instances, it
can be reduced to the study of one simple model that describes the complex scalar field
1
ϕ = √ (ϕ1 + i ϕ2 ), with the somewhat unusual potential [97, 98]
2
i
V(ϕ) = m2 ϕ∗ ϕ + λ [ϕ4 − (ϕ∗ )4 ] , (7.10.1)
2
↔ 1
The quantity jµ = −i ϕ∗ ∂µ ϕ = (ϕ1 ∂µ ϕ2 − ϕ2 ∂µ ϕ1 ) corresponds to the baryon current
2
of scalar particles in the SU(5) model, while j0 is the baryon charge density nB of the field
ϕ. The equations of motion of the fields ϕ1 and ϕ2 are
∂V
ϕ̈1 + 3 H ϕ̇1 = − = −m2 ϕ1 + 3 λ ϕ21 ϕ2 − λ ϕ32 , (7.10.2)
∂ϕ1
∂V
ϕ̈2 + 3 H ϕ̇2 = − = −m2 ϕ2 − 3 λ ϕ22 ϕ1 + λ ϕ31 . (7.10.3)
∂ϕ2
During inflation, when H is a very large quantity, the fields ϕi evolve slowly, so that
as usual the terms ϕ̈i in (7.10.2) and (7.10.3) can be neglected. This then leads to an
expression for the density nB at the time of inflation:
!
1 ∂V ∂V
nB ≡ j0 = ϕ1 − ϕ2
3H ∂ϕ2 ∂ϕ1
 157

λ
= (ϕ41 − 6 ϕ21 ϕ22 + ϕ42 ) . (7.10.4)
3H
1 2 2
If, for example, we take the initial conditions to be ϕ2 > ∼ 4 ϕ1 > 0, λ ϕi ≪ m , then we
find from (7.10.2)–(7.10.4) that during inflation the field ϕ1 evolves very slowly, and it
remains much smaller than ϕ2 , so that during the inflationary stage is nB approximately
λ 4
constant in magnitude and equal to its initial value, nB ≈ ϕ.
3H 2
To clarify the physical meaning of this result, let us write down the equation for the
partially conserved current in our model in the following form:
d(nB a3 )
!
−3 ∂V ∂V
a ≡ ṅB + 3 nB H = i ϕ∗ −ϕ ∗ , (7.10.5)
dt ∂ϕ ∂ϕ
where a(t) is the scale factor. If there were no term ∼ i λ (ϕ4 − (ϕ∗ )4 ) in (7.10.1) leading
to nonconservation of baryon charge, the total baryon charge of the universe B ∼ nB a3
would be constant, and the baryon charge density nB would become exponentially small
at the time of inflation. In our example, however, the right-hand side of (7.10.5) does
not vanish, and serves as a source of baryon charge. Bearing in mind then that all fields
vary very slowly during inflation, so that ṅB ≪ 3 nB H, (7.10.5) again implies (7.10.4), as
obtained previously.
In other words, due to the presence of the last term in (7.10.1), the baryon charge
density varies very slowly during inflation, as do the fields ϕi (see (7.10.4)), while the
total baryon charge of the part of the universe under consideration grows exponentially.
The baryon charge density and its sign depend on the initial values of the fields ϕi , and
will be different in different parts of the universe.
When the rate of expansion of the universe becomes low, the field begins to oscillate in
the vicinity of the minimum of V(ϕ) at ϕ = 0. While this is going on, the gradual decrease
in the amplitude of oscillation reduces terms ∼ λ ϕ4 in (7.10.5) which are responsible for
the nonconservation of baryon charge; the total baryon charge of the field ϕ will then
be conserved, and its density will fall off as a−3 (t). Notice that at that point the energy
m2 ϕ2
density of the scalar field ρ ∼ will also fall off as a−3 (t). This coincidence has a
2
very simple meaning. As we have already discussed, a homogeneous field ϕ oscillating
with frequency m can be represented as a coherent wave consisting of particles of the field
ρ m 2
ϕ with particle density nϕ = = ϕ , where ϕ is the amplitude of the oscillating field.
m 2
Some of these particles have baryon charge B = +1, and some have B = −1. The baryon
nB
charge density nB is thus proportional to nϕ , so that the ratio is time-independent
nϕ
and cannot be bigger than unity in absolute value:
|nB |
= const ≤ 1 . (7.10.6)
nϕ
nB
This ratio is determined by the initial conditions. The oscillatory regime sets in at
nϕ
 158

H ∼ m, so (7.10.4) implies that at that stage

nB λ ϕ̃42
≈ , (7.10.7)
nϕ 3m
where ϕ̃2 is the value of the field ϕ2 at the onset of the oscillation stage. In the realistic
SU(5) model, Eq. (7.10.7) also contains a factor cos 2 θ, where θ is the angle between
the v and a fields in the complex plane. The ϕ-particles are unstable and decay into
leptons and quarks. The temperature of the universe rises at that point, but it cannot
rise much higher than m, since at high temperature the quarks have an effective mass
mq ∼ g T ∼ T, so that the field ϕ cannot decay when T ≫ m. The field ϕ therefore
oscillates and decays gradually, rather than suddenly, and in the process it warms the
universe up to a constant temperature T ∼ m ∼ 102 GeV. By the end of this stage, the
entire baryon charge of the scalar field has been transformed into the baryon charge of
the quarks, and for every quark or antiquark produced through the decay of a ϕ-particle,
there is approximately one photon of energy E ∼ T ∼ m. This means that the density
of photons nγ produced by the decaying field ϕ is of the same order of magnitude as nϕ .
The baryon asymmetry of the universe thus engendered is
nB nB λϕ̃2 ϕ̃2
∼ ∼ cos 2 θ · 22 ∼ cos 2 θ · 22 . (7.10.8)
nϕ nγ m MX

Note that this equation is only valid when λ ϕ̃22 ≪ m, that is, when ϕ̃2 ≪ MX , so only
from that point onward can the violation of baryon charge conservation be neglected, with
nB
the quantity becoming constant. As expected from (7.10.6), the baryon asymmetry
nϕ
of the universe as given by (7.10.8) then turns out to be less than unity. However, from
(7.10.8), it follows that the mechanism of baryogenesis discussed above may even be too
nB
efficient. For example, for ϕ̃2 ∼ MX , Eq. (7.10.8) yields = O(1). We must therefore
nγ
nB
try to understand what ϕ̃2 should be equal to, and how to reduce to the desirable
nϕ
nB
value ∼ 10−9 .
nϕ
Research into this question has shown that just like money, baryon asymmetry is hard
to come by but easy to get rid of [98]. One mechanism for reducing the baryon asymmetry
is the previously cited nonperturbative scheme [129]. If, for example, the temperature of
the universe following decay of the field ϕ exceeds approximately 200 GeV, then virtually
the entire baryon asymmetry that has been produced will disappear, with the exception
of a small part resulting from processes that violate B − L invariance. This residual can
nB
in fact account for the observed asymmetry ∼ 10−9 . Another possibility is that the
nϕ
temperature is less than 200 GeV when decay of the ϕ-field ends; that is, the baryons do
not burn up, but the initial value of the field ϕ is fairly small. This could happen, for
instance, if the fields ϕi were to vanish due to high-temperature effects or interaction with
the fields responsible for inflation. The role played by these fields would then be taken
 159

H √
up by their long-wave quantum fluctuations with an amplitude proportional to Ht
2π
(see (7.3.12)), which could be several orders of magnitude less than MX .
nB
Finally, the Anthropic Principle provides one more plausible explanation of why
nγ
is so small in the observable part of the universe. The fields ϕi and the quantity cos 2 θ
take on all possible values in different regions of the universe. In most such regions, ϕ
nB
can be extremely large, and | cos 2 θ| ∼ 1. But these are regions with ≫ 10−9 , and
nγ
life of our type it is impossible. The reason is that for a given amplitude of perturbations
δρ
, elevating the baryon density by just two to three orders of magnitude results in the
ρ
formation of galaxies having extremely high matter density and a completely different
complement of stars. It is therefore not inconsistent to think that there are relatively few
regions of the universe with small initial values of ϕ and cos 2 θ — but these are just the
regions with the highest likelihood of supporting life of our type. We will discuss this
problem in a more detailed way in Chapter 10.
In addition to the mechanism discussed above, several more that could operate at
2
temperatures T < ∼ 10 GeV have recently been proposed [130, 131, 178, 261–263]. It is still
difficult to say which of these are realistic. One important point is that many ways have
been found to explain the baryon asymmetry of the universe, and superhigh temperatures
T ∼ MX ∼ 1014 –1015 GeV, which occur only after extremely efficient reheating, are not
at all mandatory. In principle, baryon asymmetry could even occur if the temperature of
the universe never exceeded 100 GeV! This then substantially facilitates the construction
of realistic models of the inflationary universe. On the other hand, the realization that it
is possible to construct a consistent theory of the evolution of the universe in which the
temperature may never exceed T ∼ 102 GeV ∼ 10−17 MP leads us yet again to ponder the
extent to which our notions have changed over the past few years, and to wonder what
surprises might await us in the future.
 8
The New Inflationary Universe Scenario

8.1 Introduction. The old inflationary universe scenario

In the previous chapter, we described the building blocks needed for a complete theory
of the inflationary universe. It is now time to demonstrate how all parts of the theory
that we have described thus far may be combined into a single scenario, implemented in
the context of some of recently developed theories of elementary particles. As we have
already pointed out, however, there are presently two significantly different fundamental
versions of inflation theory, namely the new inflationary universe scenario [54, 55], and
the chaotic inflation scenario [56, 57]. Although we lean toward the latter, in view of its
greater naturalness and simplicity, it is still too soon to render a final decision. Moreover,
many of the results obtained in the course of constructing the new inflationary universe
scenario will prove useful, even if the scenario itself is to be abandoned. We therefore
begin our exposition with a description of the various versions of the new inflationary
universe scenario, and in the next chapter we turn to a description of the chaotic inflation
scenario. Our description of the former would be incomplete, however, if we did not say
a few words about the old inflationary universe scenario proposed in the important paper
by Guth [53].
As stated in Chapter 1, the old scenario was based on the study of phase transitions
from a strongly supercooled unstable phase ϕ = 0 in grand unified theories. The theory
of such phase transitions had been worked out long before Guth’s effort (see Chapter 5),
but nobody had attempted to use that theory to resolve such cosmological problems as
the flatness of the universe or the horizon problem.
Guth drew attention to the fact that upon strong supercooling, the energy density of
relativistic particles, being proportional to T4 , becomes negligible in comparison with the
vacuum energy V(ϕ) in the vacuum state ϕ = 0. This then means that in the limit of
extreme supercooling, the energy density ρ of an expanding (and cooling) universe tends
to V(0) and ceases to depend on time. At large t, then, according to (1.3.7), the universe
expands exponentially,
a(t) ∼ eH t , (8.1.1)
where the Hubble constant at that time is
s
8 π V(0)
H= . (8.1.2)
3 MP
 161

If all of the energy is rapidly transformed into heat at the time of a phase transition to
the absolute minimum of V(ϕ), the universe will be reheated to a temperature TR ∼
[V(0)]1/4 after the transition, regardless of how long the previous expansion went on (this
circumstance was exploited earlier by Chibisov and the present author to construct a
model of the universe which could initially be cold, but would ultimately be reheated by
a strongly exoergic phase transition; this model has been reviewed in Refs. [24, 105]).
Since the temperature TR to which the universe is reheated after the phase transition
does not depend on the duration of the exponential expansion stage in the supercooled
state, the only quantity that depends on the length of that stage is the scale factor a(t),
which grows exponentially at that time. But as we have already remarked, the universe
becomes flatter and flatter during exponential expansion (inflation). This is an especially
clear-cut effect when one considers why the total entropy of the universe is so high,
87
S> ∼ 10 (as noted in Chapter 1, this problem is closely related to the flatness problem).
Prior to the phase transition, the total entropy of the universe could be fairly low.
But afterwards, it increases markedly, with
3 3 3 3/4
S>
∼ a TR ∼ a [V(0)] ,
where a3 can be exponentially large. For example, let the exponential expansion begin
in a closed universe at a time when its radius is a0 = c1 M−1 P , and the vacuum energy is
V(0) = c2 M4P , where c1 and c2 are certain constants. In realistic theories, c1 lies between
1 and 1010 , and c2 is of order 10−10 ; we will soon see that the quantity of interest depends
very weakly on c1 and c2 . Following exponential expansion lasting for a period , the total
entropy of the universe becomes
3/4
S ∼ a30 e3 H ∆t T3R ∼ c31 c2 e3 H ∆t , (8.1.3)

whereupon S exceeds 1087 if

−1 1/4
∆t >
∼ H (67 − ln c1 c2 ) . (8.1.4)
1/4
Under typical conditions, the absolute value of ln c1 c2 will be 10 at most. The implica-
tion is that in order to solve the flatness problem, it is necessary that the universe be in
a supercooled state ϕ = 0 for a period
s
−1 3
∆t >
∼ 70 H = 70 MP . (8.1.5)
8 π V(0)
It must be noted here that if ∆t is much greater than 70 H−1 (as will happen in any
realistic version of the inflationary universe scenario), then after inflation and reheating,
ρ
the universe will be almost perfectly flat, with Ω = = 1. Allowing for moderate local
ρc
variations of ρ on the scale of the observable part of the universe, this is one of the most
important observational predictions of the inflationary universe scenario.
It can readily be shown that the condition (a T)3 > ∼ 1087 means that the “radius” of
the universe a ∼ c1 M−1P after expansion up through the present epoch will exceed the size
 162

of the observable part of the universe, l ∼ 1028 cm (see the preceding chapter). But what
this means is that in a time only slightly greater (by H−1 ln c1 ) than 70 H−1, any region of
space of size ∆l ∼ M−1 P would have inflated so much that by the present epoch it would
be larger than the observable part of the universe.
If we then bear in mind that we are considering processes taking place in the post-
Planckian epoch (ρ < M4P , T < MP , t > M−1 −1
P ), it becomes clear that a region ∆l ∼ MP in
size at the onset of exponential expansion must necessarily be causally connected. Thus,
in this scenario, the entire observable part of the universe results from the inflation of a
single causally connected region, and the horizon problem is thereby solved.
The primordial monopole problem could in principle also be solved within the frame-
work of the proposed scenario. Primordial monopoles are produced only at the points of
collisions of several different bubbles of the field ϕ that are formed during the phase tran-
sition. If the phase transition is significantly delayed by supercooling, then the bubbles
will become quite large by the time they begin to fill the entire universe, and the density
of the monopoles produced in the process will be extremely low.
Unfortunately, however, as noted by Guth himself, the scenario that he had proposed
led to a number of undesirable consequences with regard to the properties of the universe
after the phase transition. Specifically, within the bubbles of the new phase, the field
ϕ rapidly approached the equilibrium field ϕ0 corresponding to the absolute minimum
of V(ϕ), and all the energy of the unstable vacuum with ϕ = 0 within the bubble was
transformed into kinetic energy of the walls, which moved away from the center of the
bubble at close to the speed of light. Reheating of the universe after the phase transition
would have to result from collisions of the bubble walls, but due to the large size of the
bubbles in this scenario, the universe after collisions between bubble walls would become
highly inhomogeneous and anisotropic, a result flatly inconsistent with the observational
data.
Despite all the problems encountered by the first version of the inflationary universe
scenario, it engendered a great deal of interest, and in the year following the publication
of Guth’s work this scenario was diligently studied and discussed by many workers in
the field. These investigations culminated in the papers by Hawking, Moss, and Stewart
[112] and Guth and Weinberg [113], where it was stated that the defects inherent in this
scenario could not be eliminated. Fortunately, the new inflationary universe scenario had
been suggested by then [54, 55]; it was not only free of some of the shortcomings of the
Guth scenario, but also held out the possibility of solving a number of other cosmological
problems enumerated in Section 1.5.

8.2 The Coleman–Weinberg SU(5) theory and the new

inflationary universe scenario (initial simplified version)
The first version of the new inflationary universe scenario was based on the study of
the phase transition with the symmetry breaking SU(5) → SU(3) × SU(2) × U(1) in the
SU(5)-symmetric Coleman–Weinberg theory (2.2.16). The theory of this phase transition
 163

0 ϕ1 3ϕ1 ϕ0 ϕ

Figure 8.1: Effective potential in the Coleman–Weinberg theory at finite temperature.

Tunneling proceeds via formation of bubbles of the field ϕ <
∼ 3 ϕ1, where V(ϕ1 , T) =
V(0, T).

is very complicated. We therefore start by giving somewhat of a simplified description of

this phase transition, so as to elucidate the general idea behind the new scenario.
First of all, we examine how the effective potential in this theory behaves with respect
to the symmetry breaking SU(5) → SU(3) ×SU(2) ×U(1) (2.2.16) at a finite temperature.
As we said in Chapter 3, symmetry is restored in gauge theories, as a rule, at high
enough temperatures. It can be shown in the present case that when T ≫ MX , the
function V(ϕ, T) in the Coleman–Weinberg theory becomes
25 g 4 ϕ4 9 M4X
!
5 ϕ 1
V(ϕ, T) = g 2 T2 ϕ2 + ln − + + c T4 . (8.2.1)
8 128 π 2 ϕ0 4 32 π 2
where c is some constant of order 10. An analysis of this expression shows that at high
enough temperature T, the only minimum of V(ϕ, T) is the one at ϕ = 0; that is, symme-
try is restored. When T ≪ MX ∼ 1014 GeV, all high-temperature corrections to V(ϕ) at
ϕ ∼ ϕ0 vanish. However, the masses of all particles in the Coleman–Weinberg theory tend
to zero as ϕ → 0, so in the neighborhood of the point ϕ = 0, Eq. (8.2.1) for V(ϕ, T) holds
for T ≪ 1014 GeV as well. This means that the point ϕ = 0 remains a local minimum of
the potential V(ϕ, T) at any temperature T, regardless of the fact that the minimum at
ϕ ≈ ϕ0 is much deeper when T ≪ MX (Fig. 8.2).
In an expanding universe, a phase transition from the local minimum at ϕ = ϕ0 to a
global minimum at ϕ = ϕ0 takes place when the typical time required for the multiple
production of bubbles with ϕ 6= 0 becomes less than the age of the universe t. Study
of this question has led many researchers to conclude that the phase transition in the
Coleman–Weinberg theory is a long, drawn-out affair that takes place only when the
temperature T of the universe has fallen to approximately Tc ∼ 106 GeV (this is not an
entirely correct statement, but for simplicity we shall temporarily assume that it is, and
return to this point in Section 8.3). It is clear, however, that at such a low temperature,
the barrier separating the minimum at ϕ = 0 from the minimum at ϕ = ϕ0 will be located
 164

at ϕ ≪ ϕ0 (see Fig. 8.2), and the bubble formation process will be governed solely by
the shape of V (ϕ, T) near ϕ = 0, rather than by the value of ϕ0 . As a result, the field ϕ
within the bubbles of the new phase formed in this way is at first very small,
12 π Tc
ϕ<
∼ 3 ϕ1 ≈ s ≪ ϕ0 , (8.2.2)
MX
g 5 ln
Tc

where the field ϕ1 (is determined by the condition V(0, T) = V(ϕ1 , T) (see Fig. 5.3).
With this value of the field, the curvature of the effective potential is relatively small.
 d2 V 
 
2 < 2 2 2
|m | = ∼ 75 g Tc ∼ 25 Tc . (8.2.3)
 
 dϕ2 
 

The field ϕ within the bubble will clearly grow to its equilibrium value ϕ ∼ ϕ0 in a time
∆t > −1 −1
∼ |m | ∼ 0.2 Tc . For most of this time, the field ϕ will remain much smaller than ϕ0 .
This means that over a period of order 0.2 T−1
c , the vacuum energy of V(ϕ, T) will remain
almost exactly equal to V(0), and consequently the part of the universe inside the bubble
will continue to expand exponentially, just as at the beginning of the phase transition.
Here we have the fundamental difference between the new inflationary universe scenario
and the scenario of Guth, in which it is assumed that exponential expansion ceases at the
moment that bubbles are formed.
When ϕ ≪ ϕ0 and MX ∼ 5 · 1014 GeV, the Hubble constant H (is given by)
s s
8π M2X 3
H= V(0) = ≈ 1010 GeV . (8.2.4)
3 M2P 2 MP π
H ∆t
In a time ∆t ∼ 0.2 T−1
c , the universe expands by a factor e , where
−1
eH ∆t ∼ e0.2 H Tc ∼ e2000 ∼ 10800 . (8.2.5)

To order of magnitude, the typical size of a bubble at the instant it is formed is T−1 c ∼
10−20 cm. After expansion, this size becomes ∼ 10800 cm, which is enormously greater
than the size of the observable part of the universe, l ∼ 10−28 cm. Thus, within the scope
of this scenario, the entire observable part of the universe should lie within a single bubble.
We therefore see no inhomogeneities that might arise from bubble wall collisions.
As in the Guth scenario, exponential expansion by a factor of more than e70 (8.2.5)
enables one to resolve the horizon and flatness problems. But more than that, it makes it
possible to explain the large-scale homogeneity and isotropy of the universe (see Chapter
7).
Since bubble sizes exceed the dimensions of the observable part of the universe, and
since monopoles and domain walls are only produced near bubble walls, there should
be not a single monopole or domain wall in the observable part of the universe, which
removes the corresponding problems discussed in Section 1.5.
 165

Note that the curvature of the effective potential (8.2.1) grows rapidly with increasing
field ϕ. The slow-growth stage of the field ϕ, which is accompanied by exponential ex-
pansion of the universe, is therefore replaced by a stage with extremely rapid attenuation
of the field ϕ to its equilibrium value ϕ = ϕ0 , where it oscillates about the minimum of
the effective potential. In the model in question, q the oscillation frequency is equal to the
mass of the Higgs field ϕ when ϕ = ϕ0 , m = V′′ (ϕ0 ) ∼ 1014 GeV. The typical period
of oscillation ∼ m−1 is evidently many orders of magnitude less than the characteristic
expansion time of the universe H−1 . In studying oscillations of the field ϕ near the point
ϕ0 , one can therefore neglect the expansion of the universe. This means that at the stage
we are considering, all of the potential energy V(0) is transformed into the energy of the
oscillating scalar field. The oscillating classical field ϕ produces Higgs bosons and vector
bosons, which quickly decay. In the end, all of the energy of the oscillating field ϕ is
transformed into the energy of relativistic particles, and the universe is reheated to a
temperature [123, 124]
TR ∼ [V(0)]1/4 ∼ 1014 GeV .
The mechanism for reheating the universe in the new scenario is thus quite different from
the corresponding mechanism in the Guth scenario.
The baryon asymmetry of the universe is produced when scalar and vector mesons
decay during the reheating of the universe [36–38]. Because of the fact that processes
taking place at that time are far from equilibrium, however, the baryon asymmetry is
produced much more efficiently in this model than in the standard hot universe theory
[123].
We see, then, that the fundamental idea behind the new inflationary universe scenario
is quite simple: it requires that symmetry breaking due to growth of the field ϕ proceed
fairly slowly at first, giving the universe a chance to inflate by a large factor, and that
in the later stages of the process, the rate of growth and oscillation frequency of the field
ϕ near the minimum of V(ϕ) be large enough to ensure that the universe is reheated
efficiently after the phase transition. This idea has been used both in a refined version of
the new scenario, which we discuss next, and in all subsequent variants of the inflationary
universe scenario.

8.3 Refinement of the new inflationary universe scenario

The description of the new inflationary universe scenario in the previous section was
oversimplified, its main drawback being our neglect of the effects of exponential expansion
of the universe on the kinetics of a phase transition. When T ≫ H ∼ 1010 GeV, such a
simplification is completely admissible, but according to the discussion in Section 8.2, the
phase transition can only begin when Tc ≪ H. In that event, high-temperature effects
exert practically no influence on the kinetics of the phase transition. Indeed, the typical
time over which bubbles might be formed at a temperature Tc must certainly be greater
than
m−1 (ϕ = 0, T = Tc ) ∼ (g Tc )−1 ≫ H−1 .
 166

But in that much time, the universe expands by a factor of approximately eH/gTc , and
the temperature falls from T = Tc practically to zero. Thus, the role of high-temperature
effects is just to place the field ϕ at the point ϕ = 0, and one can then neglect all high-
temperature effects in describing the formation of bubbles of the field ϕ and the process
by which ϕ rolls down to ϕ0 . It is necessary, however, to take account of effects related
to the rapid expansion of the universe, since at that time H ≫ T.
The resulting refinement of the scenario takes place in several steps.
1) In studying the evolution of the field ϕ in an inflationary universe, one must make
allowance for the fact that the equation of motion of the field is modified, and takes the
form
1 dV
ϕ̈ + 3 H ϕ̇ − 2 ∇2 ϕ = − . (8.3.1)
a dϕ
If the effective potential is not too steep, the ϕ̈ term in (8.3.1) can be discarded, so that
the homogeneous field ϕ satisfies the equation
1 dV
ϕ̇ = − . (8.3.2)
3 H dϕ
m2 2
In particular, (8.3.2) implies that with H = const ≫ m in a theory with V = V(0)+ ϕ ,
2
m2
!
ϕ ∼ ϕ0 exp − t , (8.3.3)
3H

m2 2
and in a theory with V = V(0) − ϕ,
2
m2
!
ϕ ∼ ϕ0 exp + t . (8.3.4)
3H

This means, in particular, that the curvature of the effective potential at ϕ = 0 need
not necessarily be zero. To solve the flatness and horizon problems, it is sufficient that
the field ϕ (as well as the quantity V(ϕ)) vary slowly over a time span ∆t > −1
∼ 70 H . In
conjunction with (8.2.4), this condition leads to the constraint

H2
|m2 | <
∼ 20 . (8.3.5)

It will also be useful to investigate the evolution of a classical field in the theory
described by
λ
V(ϕ) = V(0) − ϕ4 . (8.3.6)
4
In that case, it follows from (8.3.2) that
1 1 2λ
2
− 2 = (t − t0 ) , (8.3.7)
ϕ0 ϕ 3H
 167

where ϕ0 is the initial value of the field ϕ. This means that the field becomes infinitely
large in a finite time
3H
t − t0 = . (8.3.8)
2 λ ϕ20
If λ ϕ20 ≪ H2 , then t − t0 ≫ H−1 , and ϕ will spend most of this time span in its slow
downhill roll. It is only at the end of the interval (8.3.8) that the field quickly rolls
downward, with ϕ → ∞, in a time ∆t ∼ H−1 . For λ ϕ20 ≪ H2 , therefore, the duration of
3H
the inflationary stage in the theory (8.3.6) as the field ϕ rolls down from ϕ = ϕ0 is
2 λ ϕ20
−1
(8.3.8) (to within ∆t ∼ H ). This result will shortly prove useful.
2) Corrections to the expression (8.2.1) for V(ϕ) arise in de Sitter space. If we limit
attention, as before, to the contribution to V(ϕ) from heavy vector particles (see Chapter
2), then for small ϕ (e ϕ ≪ H), V(ϕ) takes the form [264, 265]

µ21 e2 R 2 R 3 e4 ϕ4 R
V(ϕ, R) = R+ 2
ϕ ln 2
+ 2
ln 2 + V(0, R) , (8.3.9)
2 64 π µ2 64 π µ3

where R is the curvature scalar (R = 12 H2 ), and the µi are some normalization factors
with dimensions of mass, whose magnitude is determined by the normalization conditions
imposed on V(ϕ, R). When V(ϕ) ≪ M4P , the corresponding corrections to the effective
potential V(ϕ) itself are extremely small, although they can induce significant corrections
d2 V 

2
to the quantity m = that are of order e2 H2 , and these can prevent (8.3.5) from
dϕ2 ϕ=0
being satisfied. Fortunately, there does exist a choice of normalization conditions (i.e.,
a redefinition of the Coleman–Weinberg theory in curved space) for which this does not
happen, and for which m2 remains equal to zero. We shall not pursue this problem any
further here, referring the reader to Ref. [265] for a discussion of the renormalization of
V(ϕ, R) for the Coleman–Weinberg theory in de Sitter space.
3) The most important refinement of the scenario has to do with the first stage of
growth of the field ϕ. As stated earlier, some time τ ∼ O(H−1 ) after the temperature
of the universe has dropped to T ∼ H, the temperature and effective mass of the field
ϕ at the point ϕ = 0 become exponentially small. At that time, the effective potential
H
V(ϕ) (8.2.1) in the neighborhood of interest around ϕ = 0 (with H < ∼ϕ< ∼ √λ ) can be
approximated by (8.3.6), where

25 g 4 9 M4X
!
H 1
λ≈ ln − , V(0) = . (8.3.10)
32 π 2 ϕ0 4 32 π 2

According to Eq. (8.3.8), the classical motion of the field ϕ, starting out from the point
ϕ0 = 0, would go on for an infinitely long time. As we noted in Section 7.3, however,
quantum fluctuations of the field ϕ in the inflationary universe engender long-wave fluc-
tuations in the field, and on a scale l ∼ H−1 , these look like a homogeneous classical
 168

field. Making use of (7.3.12), the rms value of this field (averaged over many independent
regions of size l > −1
∼ H ), is
H q
ϕ∼ H (t − t0 ) . (8.3.11)
2π
In the case at hand, t0 is the time at which the effective mass squared of the field ϕ at
ϕ = 0 becomes much less than H2 .
Long-wave fluctuations of the field ϕ can play the role of the initial nonzero field ϕ in
Eq. (8.3.7). Here, however, we must voice an important reservation. In different regions
of the universe, the fluctuating field ϕ will take on different values; in particular, there will
always be regions in which ϕ does not decrease at all, giving rise to a self-regenerating
inflationary universe [266, 267, 204] analogous to that of the chaotic inflation scenario
[57, 132, 133] (see Section 1.8). Further on, we shall discuss the average behavior of the
fluctuating field ϕ (8.3.11).
During the first stage of the process, fluctuating (diffusive) growth of the field ϕ takes
place more rapidly than the classical rolling:

H2 λ ϕ3 λ H2 [H (t − t0 )]3/2
ϕ̇ ∼ q ≫ ∼ √ . (8.3.12)
4π H (t − t0 ) 3H 6π 2π

This stage lasts for a time √

2
∆t = t − t0 ∼ √ , (8.3.13)
H λ
during which the mean field ϕ (8.3.11) rises to
 1/4
H 2
ϕ0 ∼ . (8.3.14)
2π λ
To a good approximation, subsequent evolution of the field ϕ may be described by Eq.
(8.3.7), where we must substitute t0 + ∆t for t0 . The overall duration of the rolling of the
field ϕ from ϕ = ϕ0 to ϕ = ∞ is
√
3H 3 2π
t − (t0 + ∆t) = = √ , (8.3.15)
2 λ ϕ20 λH
and the total duration of inflation is given by
√
4 2π
t − t0 ∼ √ . (8.3.16)
λH
During that time, the size of the universe grows by approximately a factor of
√ !
4 2π
exp(H (t − t0 )) ∼ exp √ . (8.3.17)
λ
 169

The condition H (t − t0 ) >

∼ 70 leads to the constraint [265, 128, 134, 135]
1
λ<
∼ 20 , (8.3.18)

which can also be satisfied, in principle, in the SU(5) Coleman–Weinberg theory.

With minor modifications, the foregoing discussion also applies to the case in which
m2 ≡ V′′ (0) < 0, |m2 | ≪ H2 , as well as to the case in which the effective potential has a
shallow local minimum at ϕ = 0 — that is, when 0 < m2 ≪ H2 .
In the first of these two instances, the process whereby the field rolls down from the
point ϕ = 0 is analogous to the previous situation. In the second, diffusion of the field ϕ
looks like tunneling, the theory of which was discussed in an Section 7.4.
Clearly, the details of the behavior of the scalar field ϕ as it undergoes a phase tran-
sition from the point ϕ = 0 to a minimum of V(ϕ) at ϕ = ϕ0 differ from the description
given in the preceding section. Nevertheless, most of the qualitative conclusions having to
do with the existence of an inflationary regime in the Coleman–Weinberg theory remain
valid.
Unfortunately, however, the original version of the new inflationary universe scenario,
based on the theory (8.2.1), is not entirely realistic, the point being that fluctuations of
the scalar field ϕ that are generated during the inflationary stage give rise to large density
inhomogeneities by the time inflation has ended. Specifically, according to (7.5.22), after
the inflation, reheating, and subsequent cooling of the universe, density inhomogeneities
s
δρ(ϕ) 48 2 π [V(ϕ)]3/2
= (8.3.19)
ρ 5 3 M3P V′ (ϕ)

will be produced. In this expression, ϕ is the value of the field at the time when the
corresponding fluctuations δϕ had a wavelength l ∼ k −1 ∼ H−1 . In the new inflationary
universe scenario, V(ϕ) ≈ V(0) at the time of inflation. Let us estimate the present-day
wavelength of a perturbation whose wavelength was previously l ∼ [H(ϕ)]−1 . Equation
(8.3.8) tells us that after the field becomes equal to ϕ, the universe still has an inflation
2
!
3H
factor of exp to go. Estimates analogous to those made in the previous chapter
2 λ ϕ2
then show that after inflation and the subsequent stage of hot universe expansion, the
wavelength l ∼ [H(ϕ)]−1 typically increases to

3 H2
!
l ∼ exp cm . (8.3.20)
2 λ ϕ2

From (8.3.19) and (8.3.20), we obtain

√
δρ 9 H3 2 6√
∼ ∼ λ ln3/2 l [cm] , (8.3.21)
ρ 5 π λ ϕ3 5π
just as in the chaotic inflation scenario (7.5.29). At the galactic scale lg ∼ 1022 cm,
 170

δρ √
∼ 110 λ . (8.3.22)
ρ
δρ
This means that ∼ 10−5 when
ρ
λ ∼ 10−14 , (8.3.23)
again as in the chaotic inflation scenario; see (7.5.38). In the original version of the
new inflationary universe scenario, the condition (8.3.23) was not satisfied. This made
it necessary to seek other more realistic models in which the new inflationary universe
scenario might be realizable, and we now turn to a discussion of the models proposed.

8.4 Primordial inflation in N = 1 supergravity

The main reason why the new inflationary universe scenario has not been fully imple-
mented in the Coleman–Weinberg SU(5) theory is that the scalar field interacts with
vector particles, and as a result it acquires an effective coupling constant λ ∼ g 4 ≫ 10−14 .
The conclusion to be drawn, then, is that the (inflaton) field ϕ responsible for the inflation
of the universe must interact both with itself and with other fields extremely weakly. In
particular, it must not interact with vector fields, or in other words it ought to be a singlet
under gauge transformations in grand unified theories.
A long list of requirements has been formulated which must be satisfied in order for
a theory to provide a feasible setting for the new inflationary universe scenario [268].
Specifically, the effective potential at small ϕ must be extremely flat (as can be seen
from (8.3.5) and (8.3.23)), and near its minimum at ϕ = ϕ0 , it must be steep enough to
ensure efficient reheating of the universe. Next, after the main requirements for a theory
have been formulated, the search for a realistic elementary particle theory of the desired
type begins. Since the step following the construction of grand unified theories was the
development of phenomenological theories based on N = 1 supergravity, there has been a
great deal of work attempting to describe inflation within the scope of these theories (for
example, see [269]–[271]).
In N = 1 supergravity, the inflaton field ϕ responsible for inflation of the universe is
represented by the scalar component z of an additional singlet of the chiral superfield Σ.
In the theories considered, the Lagrangian for this field can be put into the form [272]
L = Gzz ∗ ∂µ z ∂ µ z ∗ − V(z, z ∗ ) , (8.4.1)
V(z, z ∗ ) = eG (Gz G−1zz ∗ Gz ∗ − 3) , (8.4.2)

where G is an arbitrary real-valued function of z and z ∗ , Gz is its derivative with respect to

z, and Gzz ∗ is its derivative with respect to z and z ∗ . In the minimal versions of the theory,
one imposes the constraint Gzz ∗ = 1/2 on G so that the kinetic term in (8.4.1) takes on the
standard (minimal) form ∂µ z ∂ µ z ∗ (up to a factor 1/2), while the form chosen for G itself is
z z∗
G(z, z ∗ ) = + ln |g(z)|2 , (8.4.3)
2
 171

where g(z) is an arbitrary function of the field z, called the superpotential; all dimensional
MP
terms in (8.4.3) are expressed in units of √ . The effective potential is then given by
8π
  2 
z ∗ /2
 dg z2 
V(z, z ∗ ) = ez 2  + g − 3 |g 2| . (8.4.4)

 dz 2 

The function g is subject to two constraints, namely V(z0 ) = 0 and g(z0 ) ≪ 1, where
z0 is the point at which V(z, z ∗ ) has its minimum. The first condition means that the
vacuum energy vanishes at the minimum of V(z, z ∗ ), and the second is required in order
that the mass of the gravitino m3/2 , which is proportional to g(z0 ), be much lower than
the other masses that occur in the theory. This requirement is necessary for the solution
of the mass hierarchy problem in the context of N = 1 supergravity [15].
The superpotential g(z) can be expressed as a product µ3 f (z), where m is some
parameter with dimensions of mass. The potential V(z, z ∗ ), and thus the effective cou-
pling constants for the z and z ∗ fields, are consequently proportional to µ6 . The choice
µ ∼ 10−2–10−3 , which seems to be a fairly natural one, therefore results in the appear-
ance of extremely small effective coupling constants λ ∼ 10−12 –10−18 , which is just what
δρ
is needed to obtain the desired amplitude ∼ 10−4 –10−5 if inflation takes place in the
ρ
theory (8.4.4). This variant of the new inflationary universe scenario was called the pri-
mordial inflation scenario by its authors [270], since it was expected to be played out on
an energy scale far exceeding that of the grand unified theories. In fact, it has turned out
that the corresponding energy scales are virtually identical.
The development of the primordial inflation scenario was party to many interesting
ideas and considerable ingenuity. Unfortunately, however, no realistic versions of this
scenario (or indeed of any other versions of the new inflationary universe scenario) have yet
been suggested. The principal reason for this is that particles of the field z, interacting very
weakly (either gravitationally or through a coupling constant µ6 ∼ 10−14 ) with one another
and with other fields, were not in a state of thermodynamic equilibrium in the early
universe. Furthermore, even if they had been, the corresponding corrections of the type
λ z z ∗ T2 to V(z, z ∗ ) are so small that they are incapable of changing the initial value of the
field z; that is, in most models of this kind, they cannot raise the field z to a maximum of
the potential V(z, z ∗ ), as required for the onset of inflation in this scenario [115, 116] (this
point will be treated in more detail in Section 8.5). Meanwhile, as we shall show in Chapter
9, the chaotic inflation scenario can be implemented in N = 1 supergravity [273, 274].

8.5 The Shafi–Vilenkin model

It was Shafi and Vilenkin who came closest to a consistent implementation of the new
inflationary universe scenario [275] (see also [276]). They returned to a consideration of
the SU(5)-symmetric theory of Coleman and Weinberg, with symmetry breaking due to
 172

the Coleman1 Weinberg mechanism occurring not in the field Φ, which interacts with
vector bosons through a gauge coupling constant g, but in a new, specially introduced
field χ, an SU(5) singlet, which interacts very weakly with the superheavy Φ and H5 Higgs
fields. The effective potential in this model is
1 1 γ +
V = a Tr(Φ2 )2 + b Tr Φ4 − α (H+ 2
5 H5 ) Tr Φ + (H H )2
4 2 4 5 5
λ1 4 λ2 2 λ3 2 +
− β H+ 2
5 Φ H5 + χ − χ Tr Φ2 + χ H5 H5
4! 2 2
χ2
+ A χ4 ln 2 + C + V(0) . (8.5.1)
χ0

where a, b, α, and γ are all proportional to g 2; C is some normalization constant; 0 <

λi ≪ g 2, λ1 ≪ λ22 , λ23 , and the magnitude of A is determined by radiative corrections
associated with the interaction of the field χ with the fields Φ, H5 , and (indirectly) with
the X and Y vector mesons.
In the present case, it is not an entirely trivial matter to calculate A, and the procedure
requires some explanation. Spontaneous SU(5) symmetry breaking takes place when the
1
nonzero classical field χ emerges, thanks to the term − λ2 χ2 Tr Φ2 in (8.5.1). The
2
symmetry breaks down to SU(3) × SU(2) × U(1) by virtue of the emergence of the field
s
2 3 3
 
Φ= ϕ · diag 1, 1, 1, − , −
15 2 2
(see (1.1.19)), where
2 λ2 2
ϕ2 = χ (8.5.2)
λc
7
and λc = a + b. The time needed for the field ϕ to grow to the value (8.5.2) is then
√ 15
τ ∼ ( λ2 χ)−1 , which is much less than the typical time scale of variations in χ at the
time of inflation (see below). The field ϕ thus continuously follows the behavior of the
field χ. Consequently, not only does a change in χ alter the masses of those particles
with which this field interacts directly (such as Φ and H5 ), but also those that interact
with the field ϕ — in particular, the X and Y vector mesons. Here the behavior of the
H5 boson masses is especially interesting. The first two components of H5 play the role of
the Higgs field doublet in SU(2) × U(1) symmetry breaking. These should be quite light,
with m2 ∼ 102 GeV ≪ m3 , MX , MY ,. . . . To lowest order, one may therefore put m2 = 0
at the minimum of V(ϕ, χ).
The general expression for the doublet and triplet masses of the H fields follows from
(8.5.1):

m22 = λ3 χ2 − (α + 0.3 β) ϕ2 , (8.5.3)

β
m23 = m22 + ϕ2 . (8.5.4)
6
 173

Making use of (8.5.2), one obtains

2 λ2
λ3 = (α + 0.3 β) . (8.5.5)
λc
This implies that not just at the minimum of V(ϕ, χ), but anywhere along a trajectory
along which the field χ varies,
β 2
m22 = 0, m23 = ϕ . (8.5.6)
6
β 2
The constant λ3 is thus not independent, and the value of m23 is proportional to ϕ ,
6
rather than λ3 χ2 . A calculation of the radiative corrections to V(ϕ, χ) in the vicinity of
a trajectory down which the field χ is rolling, with λi , β ≪ g 2 , finally gives [277]

λ22 25 g 4 14 b2
!
A= 1+ + . (8.5.7)
16 π 2 16 λ2c 9 λ2c

(This expression differs slightly from the one given in Ref. [275].) If one takes for simplicity
a ∼ b ∼ g 2 , then Eq. (8.5.7) yields

A ∼ 1.5 · 10−2 λ22 . (8.5.8)

The effective potential V(ϕ, χ) in the theory (8.5.1) looks like

λc 4 λ2 2 2 λ1 4 χ
 
V= ϕ − ϕ χ + χ + A χ4 ln + C + V(0) , (8.5.9)
16 4 4 M
where M and C are some normalization parameters. To determine M, C, and V(0), one
should use Eq. (8.5.2):

λ2 χ
 
V = − 2 χ4 + A χ4 ln + C + V(0) . (8.5.10)
4 λc M
With an appropriate choice of the normalization constant C, the effective potential (8.5.10)
can be put in the standard form

A χ40
!
4 χ 1
V(χ) = A χ ln − + , (8.5.11)
χ0 4 4

where χ0 gives thesposition of the minimum of V(χ). The corresponding minimum in ϕ

2 λ2
is located at ϕ0 = χ0 (see (8.5.2)), and the mass of the X boson is equal to
λc
s
5 g ϕ0
MX = ∼ 1014 GeV .
3 2
 174

Hence, s
MX 6 λc
χ0 ∼ ,
g 5 λ2
and
A 4
χ ≈ M4X .
V(0) =
4 0
The high-temperature correction to the effective potential (8.5.11) is given by
5
 
∆V(χ, T) = λ3 − λ2 T2 χ2 , (8.5.12)
12
12
which for λ3 > λ2 could lead to the restoration of symmetry, χ → 0 (see the next
5
section, however). Upon cooling, the process of inflation would begin, which would be
very similar to the process described in Section 8.3.
To determine! the numerical value of the parameter A, one should first determine the
χ0
value of ln at which the observable structure of the universe is actually formed —
χ
this occurs when a time t ∼ 60 H−1 remains prior to the end of inflation. According to
(8.3.8), the magnitude of the field χ at that point is given by
H2
χ2 ∼ (8.5.13)
40 λ(χ)
!
χ
where for ln ≫ 1, the effective coupling constant λ(χ) and the Hubble constant H
χ0
are
!
χ0
λ(χ) ≈ 4 A ln ∼ 10−14 ,
χ
v
u 8 π V(0) M2X
u
H = t ∼3 ∼ 3 · 109 GeV , (8.5.14)
3 M2P MP

(see (8.3.23)), whereupon

χ ∼ 5 · 1015 GeV . (8.5.15)
!
χ0
Inserting these values, ln is found to be of order 3 (see below). From (8.5.8) and
χ
(8.5.14), it follows that
λ2 ∼ 3 · 10−6 , (8.5.16)
s
MX 6 λc
χ0 ∼ ∼ 1017 GeV (8.5.17)
g 5 λ2
12 λ2
According to (8.5.11), the value of λ3 should be greater than . But λ3 cannot be
5
much greater than λ2 , since it can be shown that if it were, ϕ would not vanish at high
 175

temperature [275]. In accordance with [275], we shall therefore assume that λ3 ∼ 3 · 10−6 ,
like λ2 .
We have from (8.3.17) that in the present model, a typical inflation factor is of order
 √ 
4 2π 8
exp  q ∼ 1010 , (8.5.18)
λ(χ)

which is more than adequate.

Unfortunately, both reheating and the baryon asymmetry production in the post-
inflation universe are rather inefficient in this model. After inflation, the field χ oscillates
about the minimum of V(χ) at χ = χ0 at a very low frequency,
√
mχ = 22 A χ0 ∼ 1011 GeV . (8.5.19)

The principal decay mode of the field χ is χ χ → H+ 3 H3 , where H3 is the triplet of heavy
Higgs bosons. Subsequent decay of the H3 bosons gives rise to the baryon asymmetry of
the universe. The corresponding part of the effective Lagrangian responsible for decay of
β λ2 2 +
the field χ is of the form χ H3 H3 . But such a process is only possible if m3 < mχ ∼
6 λc
1011 GeV, and if H3 had such a mass, the proton lifetime would be unacceptably short,
making the entire scheme unrealistic.
Let us digress from this problem for a moment, since in any case the SU(5) model
in question is in need of modification — it gives a high probability of proton decay even
when m3 ≫ mχ . In order for the decay χ χ → H+ 3 H3 to occur, let us take mχ ∼ mH+ ,
3
that is, β ∼ 10−6 . In that event,

(10−11 χ)2
Γ(χ χ → H+
3 H3 ) ∼ · O(10−2 ) ∼ 10−2 GeV , (8.5.20)
mχ

and so, according to (7.9.10),

q
TR ∼ 10 −1
Γ MP ∼ 3 · 107 GeV . (8.5.21)

Baryon asymmetry formation is a possible process in this model, since H3 bosons are cre-
ated and destroyed, but each decay generates
m3
 
3
O ∼ 3 · 10 photons of energy E ∼ T. This tends to make the occurrence of a
TR
baryon asymmetry less likely by a factor of 3 · 103 . To circumvent this problem, one
should either invoke alternative baryon production mechanisms (see Chapter 7), or mod-
ify the Shafi–Vilenkin model. We shall return to this question in the next chapter; for
the time being, let us try to analyze the main results obtained above, and evaluate the
prospects for further development of the new inflationary universe scenario.
 176

8.6 The new inflationary universe scenario: problems and prospects

As we have already seen, the basic problems with the new inflationary universe scenario
have to do with the need to obtain small density inhomogeneities in the observable part
of the universe after inflation. This question deserves more detailed discussion.
1. As we mentioned in Section 7.5, the condition A < −4
∼ 10 imposes an overall con-
straint on V(ϕ) in the new inflationary universe scenario,
−10 4
V(ϕ) <
∼ 10 MP . (8.6.1)

This means that in any version of this scenario, including the primordial inflation scenario,
the process of inflation can only begin at a time
s
−1 3 M2P
t>
∼H ∼ ∼ 10−36 sec ,
8πV
or in other words, six orders of magnitude later than the Planck time tP ∼ M−1 P ∼ 10
−43

sec. Bearing in mind, then, that a typical total lifetime for a hot, closed universe is of
order t ∼ tP (see Chapter 1), it is clearly almost always the case that a closed universe
simply fails to survive until the beginning of inflation; i.e., the flatness problem cannot
be solved for a closed universe. The new inflationary universe scenario can therefore only
be realized in a topologically nontrivial or a noncompact (infinite) universe, and only in
5 −1
those parts of the latter which don’t collapse and are sufficiently large (l > ∼ 10 MP ) at
the moment when the matter density therein becomes less than V(ϕ) ∼ 10−10 M4P .
2. Let us now direct our attention to the fact that the new inflationary universe
scenario can only be realized in theories in which the potential energy V(ϕ) takes on a
highly specific form where, as we have seen, the coupling constants are strongly interre-
lated. Considerable ingenuity is required to construct such theories, with the result that
the original simplicity underlying the idea of inflation of the universe is gradually lost in
the profusion of conditions and reservations required for its implementation.
3. The basic difficulty of the new inflationary universe scenario relates to the question
of how the field ϕ reaches the maximum of the effective potential V(ϕ) at ϕ = 0. This
problem turned out to be an especially serious one as soon as it was realized that the field
ϕ ought to interact extremely weakly with other fields.
To get to the heart of the problem, let us examine some region of a hot universe in
which the field ϕ has the initial value ϕ ∼ ϕ0 . Suppose that high-temperature corrections
lead to a correction to V(ϕ) of the form

α2 2 2
∆V ∼ ϕ T . (8.6.2)
2
H−1
The age of the hot universe equals (1.4.6):
2
s s
H−1 MP 3 MP 3 MP
t= = < ∼ . (8.6.3)
2 2 8πρ 2 8 π ∆V 4αϕT
 177

Over this time span, the corrections (8.6.2) can only change the initial value ϕ = ϕ0 if the
typical time τ = (∆m)−1 (T) ∼ (α T)−1 is less than the age of the universe t, whereupon
we obtain the condition
MP
ϕ0 <∼ 3 . (8.6.4)
High-temperature corrections can thus affect the initial value of the field ϕ only if the
MP
latter is less than . Meanwhile, theories with V(ϕ) ∼ ϕn place no constraints on
3
4
the initial value of the field ϕ except to require that V(ϕ) < ∼ MP . For example, in the
λ ϕ4 4
Shafi–Vilenkin theory (as in a theory with λ ∼ 10−14 ), the constraint V(χ) <∼ MP
4
implies that the field χ can initially take on any value in the range

−104 MP <
∼χ<
4
∼ 10 MP , (8.6.5)

and only less than 10−4 of this interval is pertinent to values of χ for which high-
temperature corrections can play any role.
MP
For the case ϕ < ∼ 3 one can make another estimate. In a hot universe with N
different particle species, s
1 45 MP
t<
∼ 4 π π N T2 , (8.6.6)

(see (1.3.20)). Comparing t from (8.6.6) with τ ∼ (α T)−1 , we see that high-temperature
effects only begin to change the field ϕ at a temperature
√
α MP 45 α MP
T< ∼ T1 ∼ 4 π √π N ∼ 50 , (8.6.7)

when the overall energy density of hot matter is

π2 3 · 10−3 α4 M4P
ρ(T1 ) ∼ N T41 <
∼ α 4
M4
P ∼ 2
∼ 10−7 α4 M4P (8.6.8)
30 N
for N ∼ 200 (as would be the case in grand unified theories). Notice, however, that
the process whereby the field ϕ decreases can continue only so long as the total density
ρ(T) is not comparable to V(0), since soon afterward, high-temperature effects become
exponentially small due to inflation. This leads to the constraint

10−7 α4 M4P > V(0) . (8.6.9)

In the theory of a field ϕ that interacts only with itself (with a coupling constant
λ ∼ 10−14 ), and in primordial inflation models, the parameter α2 is of order 10−14 , so
that (8.6.9) then becomes
−35 4
V(0) <
∼ 10 MP . (8.6.10)
This value is much smaller than the actual value of V(0) in all realistic models of new
inflation.
 178

The situation is somewhat better in the Shafi–Vilenkin model.

2 −7
There, α is of order 10 , and (8.6.9) remains valid. We must ascertain, however, whether
high-temperature effects can make χ smaller than 5 · 1015 GeV (8.5.15), which would be
necessary for the universe to inflate by a factor of e60 –e70 .
For this to happen, the field χ must be reduced to 5 · 1015 GeV at the time when the
quantity
d∆V(χ, T)
≈ α2 T2 χ
dχ
becomes less than
dV(χ) χ0
∼ 4 A χ3 ln .
dχ χ
This occurs at a temperature
T2 ∼ 1012 GeV . (8.6.11)
While the temperature T drops from T1 to T2 , the field χ oscillates in the potential
α2 2 2
∆V(χ, T) ∼ χ T at a frequency mχ ∼ α T. The rate of production of pairs by this
2
field is very low (8.5.20), so that its oscillation amplitude in the early universe decreases
mainly by virtue of the expansion of the universe. It is readily shown that in the present
α2 2 2
case (with ∆V(χ, T) ∼ χ T ), the falloff in the field χ is proportional to the temper-
2
ature T. As the temperature drops from T1 ∼ 1014 GeV to T2 ∼ 1012 GeV, the original
amplitude of the field χ is reduced by a factor of 102 , and it becomes less than ∼ 5 · 1015
GeV only if the initial field χ was less than 5 · 1017 GeV.
Thus, in order to implement the new inflationary universe scenario in the Shafi–
Vilenkin model, the field χ must originally be a factor of 20 less than MP , a requirement
that is quite unnatural.
It must be understood that the foregoing estimates are model-dependent. There do
exist theories in which the effective potential V(ϕ) rises so rapidly with increasing field ϕ
that it becomes greater than M4P for ϕ > ∼ MP . In that event, the condition ϕ0 < ∼ MP may
be warranted. In general, it is possible to suggest mechanisms whereby a field ϕ < ∼ MP
in the early universe rapidly drops to ϕ ≪ MP . But the examples considered above
show that it is indeed difficult to obtain consistency among all the requirements needed
for a successful implementation of the new inflationary universe scenario. As a result, a
consistent implementation of this scenario within the framework of a realistic elementary
particle theory is still lacking.
Of course, one cannot rule out the possibility that some future elementary particle
theory will automatically satisfy all the necessary conditions. But for now, there is no
need to insist that all of these conditions be satisfied, as there is another scenario amenable
to realization over a much wider class of theories, namely the chaotic inflation scenario.
 9
The Chaotic Inflation Scenario

9.1 Introduction. Basic features of the scenario.

The question of initial conditions
The general underpinnings of the chaotic inflation scenario were described in some detail
in Chapter 1. Rather than reiterating what has already been said, we shall attempt here
to review the basic features of this scenario, which may perhaps stand out in better relief
against the backdrop of the preceding discussion of the new inflationary universe scenario.
The basic idea behind this scenario is simply that one need no longer assume that the
field ϕ lies at a minimum of its effective potential V(ϕ) or V(ϕ, T) from the very outset in
the early universe. Instead, it is only necessary to study the evolution of ϕ for a variety
of fairly natural initial conditions, and check to see whether or not inflation sets in.
If one requires in addition that a solution to the flatness problem be accessible within
the context of this scenario even when the universe is closed, then it becomes necessary
that inflation be able to start with V(ϕ) ∼ M4P . As demonstrated in Chapter 1, this
requirement is satisfied by a broad class of theories in which the effective potential V(ϕ)
increases no faster than a power of the field ϕ in the limit ϕ ≫ M P . In principle, inflation
ϕ
can also take place in theories for which V(ϕ) ∼ exp α when ϕ ≫ MP if α is
MP
sufficiently small α <∼ 5. The general criterion for the onset of inflation follows from the
8πV
condition Ḣ ≪ H2 = and (1.7.16):
3 M2P
√
d ln V 4 π
≪ . (9.1.1)
dϕ MP
As we stated in Chapter 1, the most natural initial conditions for the field ϕ on a scale
l ∼ H−1 ∼ M−1 0 i 4
P are that ∂0 ϕ ∂ ϕ ∼ ∂i ϕ ∂ ϕ ∼ V(ϕ) ∼ MP . The probability of formation
of an inflationary region of the universe then becomes significant — we might estimate
it to be perhaps 1/2 or 1/10. For our purposes, the only important consideration is that
the probability is not reduced by a factor like exp(−1/λ) [118]. Meanwhile, it has been
argued by some authors (see [258, 278], for example) that the probability of inflation
λ ϕ4
in the theory might actually be suppressed by a factor of this kind. A thorough
4
investigation of this question is necessary for a proper understanding of those changes
 180

that the inflationary universe scenario has introduced into our conception of the world
about us. We will discuss this question here, following the arguments put forth in [118].
1. First, let us try to understand whether it is actually necessary to assume that
ϕ̇
≪ V(ϕ) from the very outset. For simplicity, we consider Eqs. (1.7.12) and (1.7.13) in
2
a flat universe (k = 0) with a uniform field ϕ, and with ϕ̇2 ≫ V(ϕ). Then (1.7.12) and
(1.7.13) imply that ϕ̈ ≫ V′ (ϕ) and
√
2 3π 2
ϕ̈ = ϕ̇ , (9.1.2)
MP
so that
√ !−1
3π 2
ϕ̇ = −|ϕ̇0 | 1 + |ϕ̇0 | t , (9.1.3)
MP
√ !
MP 2 3π
ϕ = ϕ0 − √ ln 1 + |ϕ̇0 | t . (9.1.4)
2 3π MP

−1 MP
This means that when t >
∼H ∼ , the kinetic energy of the field ϕ falls off according
|ϕ̇0 |
to a power law, as ϕ̇2 ∼ t−2 , while the magnitude of the field itself (and thus of V(ϕ) ∼ ϕ′′ )
falls off only logarithmically. The kinetic energy of the field ϕ therefore drops rapidly, and
after a short time (just several times the value of H−1 ), the field enters the asymptotic
regime ϕ̇2 ≪ V(ϕ) [118, 110].
The thrust of this result, a more general form of which was derived in [279, 280]
for both an open and closed universe, is quite simple. When ϕ̇2 > V(ϕ), the energy-
momentum tensor has the same form as the energy-momentum tensor of matter whose
equation of state is p = ρ. The energy density of such matter rapidly decreases as the
universe expands, while the value of a sufficiently flat potential V(ϕ) changes very slowly.
Let us estimate the fraction of initial values of ϕ̇ for which the universe fails to enter
λ ϕ4
the inflationary regime in the theory. This requires that ϕ̇2 remains larger than V(ϕ)
4
MP ϕ̇2
until ϕ becomes smaller than . The initial value of is of order M4P (prior to that
3 2
point, it is impossible to describe the universe classically), and the initial value ϕ0 of the
field ϕ can take on any value in the range −λ1/4 MP < ∼ϕ<
1/4
∼ λ MP . In that event, we
see from (9.1.4) that the total√ time ! needed for the field ϕ to decrease from ϕ0 to ϕ ∼ MP
1 2 3π ϕ0
is of order √ exp . In this time span, ϕ̇ is reduced in magnitude by
2 6πMP M√P
!
2 3 π ϕ0
approximately a factor exp . We then find that when λ ≪ 1, ϕ̇2 can remain
MP
larger than V(ϕ) during the whole process only if ϕ0 ∼ MP . The probability that the
field ϕ, which initially can take any value in the range from −λ1/4 MP to λ1/4 MP , winds
up being of order MP can be estimated to be λ1/4 ∼ 3 · 10−4 for λ ∼ 10−14 . It is therefore
practically inevitable that a homogeneous, flat universe passes through the inflationary
 181

λ ϕ4
regime in the theory [280, 110, 118]. One comes to the same conclusion for an open
4
universe as well. For a closed universe, the corresponding probability is of order 1/4 [280].
The reason that the probability is smaller for a closed universe is that it may collapse
before ϕ̇2 becomes smaller than V(ϕ). In any event, the probability of occurrence of an
inflationary regime turns out to be quite significant, as we expected.
2. Next, let us discuss a more general situation in which the field ϕ is inhomogeneous.
If the universe is closed, then its overall initial size l is O(M−1 −1
P ) (when l ≪ MP , the
universe cannot be described in terms of classical space-time, and in particular one cannot
say that its size l is much less than M−1 0 i
P ). If ∂0 ϕ ∂ ϕ and ∂i ϕ ∂ϕ are both severalfold
smaller than V(ϕ), which is not improbable, then the universe begins to inflate, and the
gradients ∂i ϕ soon become exponentially small. Thus, the probability of formation of a
closed inflationary universe remains almost as large as in the case considered above, even
when possible inhomogeneity of the field ϕ is taken into account.
If the universe is infinite, then the probability that the conditions necessary for inflation
actually come to pass would seem at first glance to be extremely low [258]. In fact,
λ ϕ4
if a typical initial value of the field ϕ in a theory is, as we have said, of order
4
ϕ0 ∼ λ−1/4 MP ∼ 3000 MP, then the condition ∂i ϕ ∂ i ϕ < 4
∼ MP might lead one to conclude
that ϕ should remain larger than ∼ λ −1/4
MP on a scale l > ∼ λ
−1/4
MP ∼ 3000 M−1 P .
−1
But this is highly improbable, since initially (i.e., at the Planck time tP ∼ MP ) there
can be no correlation whatever between values of the field ϕ in different regions of the
universe separated from one another by distances greater than M−1 P . The existence of such
correlation would violate causality (see the discussion of the horizon problem in Chapter
1).
The response to this objection is very simple [118, 78, 79]. We have absolutely no
reason to expect that the overall energy density ρ will simultaneously become less than
the Planck energy M4P in all causally disconnected regions of an infinite universe, since
that would imply the existence of an acausal correlation between values of ρ in different
domains of size O(M−1 P ). Each such domain looks like an isolated island of classical
space-time, which emerges from the space-time foam independently of other such islands.
During inflation, each of these islands acquires dimensions many orders of magnitude
larger than the size of the observable part of the universe. If some gradually join up with
others through connecting necks of classical space-time, then in the final analysis, the
universe as a whole will begin to look like a cluster (or several independent clusters) of
topologically connected mini-universes. However, such a structure may only come into
being later (see Chapter 10 in this regard), and a typical initial size of a domain of classical
4 −1
space-time with ρ < ∼ MP is extremely small — of the order of the Planck length lP ∼ MP .
Outside each of these domains the condition ∂i ϕ ∂ i ϕ < 4
∼ MP no longer holds, and there
is no correlation at all between values of the field ϕ in different disconnected regions of
classical space-time of size O(M−1 P ). But such correlation is not really necessary for the
realization of the inflationary universe scenario — according to the “no hair” theorem
for de Sitter space, a sufficient condition for the existence of an inflationary region of the
 182

universe is that inflation take place inside a region whose size is of order H−1 ∼ M−1 P ,
which in our example is actually the case.
We wish to emphasize once again (and this will subsequently be of some importance)
that the confusion that we have analyzed above, involving the correlation between values
of the field ϕ in different causally disconnected regions of the universe, is rooted in the
familiar notion of a universe that is instantaneously created from a singular state with
ρ → ∞, and instantaneously passes through a state with the Planck density ρ ∼ M4P .
The lack of justification for such a notion is the very essence of the horizon problem; see
Section 1.5. Now, having disposed of the horizon problem with the aid of the inflationary
universe scenario, we may possibly manage to familiarize ourselves with a different picture
of the universe, a picture whose specific features are gradually becoming clear. We shall
return to this question in the next chapter.
Evidently, the condition ∂i ϕ ∂ i ϕ < V(ϕ) invoked above can also be relaxed, in the
same way that we relaxed the requirement ∂0 ϕ ∂ 0 ϕ < V(ϕ). The basic idea here is that
if the effective potential V(ϕ) is flat enough, then during the expansion of the universe
(in all regions which neither drop out of the general expansion process nor collapse), the
gradients of the field ϕ fall off rapidly, whereas the mean value of ϕ decreases relatively
ϕ̇2
slowly. The net result is that just as in the case of kinetic energy , we ought to arrive
2
at a situation in which the energy density associated with gradients of the field ϕ in a
significant part of the universe will have fallen to much less than V(ϕ); in other words,
the conditions necessary for inflation appear. We shall not discuss this possibility any
further, as the results obtained above suffice for our purposes.
To conclude this section, let us note that the foregoing question involving acausal
correlation does not arise in realistic theories, in general, where apart from a “light” field
ϕ with λ ∼ 10−14 , there is at least one scalar field Φ with a bigger coupling constant
λΦ > −2
∼ 10 . In such theories, the “acausal correlation length” between values of Φ in
different regions is only marginally bigger than the distance to the horizon, so that even
if the aforementioned arguments concerning the acausal correlation between values of Φ
were true, the probability that inflation would be driven by the field Φ would not be
noticeably suppressed. As demonstrated in [281], long-wave fluctuations of the light field
ϕ that are generated at the time of inflation bring about self-regenerating inflationary
4
regions (see Section 1.8) filled with a quasihomogeneous field ϕ for which V(ϕ) < ∼ MP .
The heavy field Φ rapidly decreases in these regions, so that the last stages of inflation
are governed by the field ϕ with λ ∼ 10−14 , as before.
Our principal conclusion, then, is that there exists a broad class of elementary particle
theories within the scope of which inflation sets in under natural initial conditions.

9.2 The simplest model based on the SU(5) theory

The chaotic inflation scenario can be implemented within the framework of many models
(and in particular the Shafi–Vilenkin model, which was originally designed to implement
 183

the new inflationary universe scenario). But one can achieve the same end using simpler
models, since we no longer need to satisfy the numerous constraints imposed on the theory
by the new inflationary universe scenario. To be specific, there is no need to invoke the
Coleman–Weinberg mechanism; the superheavy Φ and H5 field sector in SU(5) theory can
be cast in a standard form; we can omit interactions between the χ-field and Φ-fields, and
so on.
Consider, for example, a theory with the effective potential
1 1 M2
V = a Tr(Φ2 )2 + b Tr Φ4 − Φ Tr Φ2 − a (H+
5 H5 ) Tr Φ
2
4 2 2
λ +
+ (H H )2 − β H+ 2 2 +
5 Φ H5 + m5 H5 H5
4 5 5
m2 2 λ1 4 λ2 2 +
− χ + χ + χ H5 H5 , (9.2.1)
2 4 2
and assume that a ∼ b ∼ α ∼ g 2, λ1 ≫ λ22 so that quantum corrections to λ1 may be
neglected. In this theory, in contrast to Eq. (8.5.3) and (8.5.4), the masses m2 and m3
are given by

m22 = m25 + λ2 χ2 − (α + 0.3 β) ϕ2 , (9.2.2)

β
m23 = m22 + ϕ2 . (9.2.3)
6
−1/4
Inflation takes place during the time that the χ-field rolls down from χ ∼ λ1 MP to
m
the minimum of V(χ) at χ0 = √ . We will assume for simplicity that χ0 < ∼ MP ; then
λ1
δρ
∼ 10−5 for λ1 ∼ 10−14 . The universe is reheated much more efficiently than in the
ρ
Shafi–Vilenkin model, as the terms in the Lagrangian responsible for decay of the field χ
are now of the form ∼ λ2 χ2 H+ 5 H5 (there is no additional coefficient β ∼ 10
−6
resulting
from the simultaneity of oscillations of the ϕ and χ fields). This effect and additional
energy transfer during oscillations of the H1 and H2 fields (which result from sign changes
in m22 as the field χ oscillates in the neighborhood of χ0 ) lead to rapid reheating of the
universe. This is also facilitated by an increase in oscillation frequency of the field χ.
If, for example, one takes√m ∼ 1012 GeV, then χ0 ∼ MP . The oscillation frequency of
the field χ then becomes 2 m = 1.5 · 1012 GeV. The reheating temperature TR in this
model can reach 1012 –1013 GeV. The decay χ χ → H+ 3 H3 takes place at m3 <
12
∼ 10 GeV;
the particular difficulties with proton decay which are related to the low mass of the m3
do not appear in this model, and the temperature TR is large enough to facilitate the
standard baryogenesis mechanism based on the decay of H3 particles.
The model presented here admits of a great many generalizations. For example,
m2 2 λ1 4
one can delete the terms − χ and χ from (9.2.1), leaving only the last term
2 4
λ22 χ4
!
λ2 2 + χ 1
χ H5 H5 . Then due to radiative corrections, an induced term like C ln −
2 64 π 2 χ0 4
 184

will be responsible for inflation. When λ2 ∼ 10−6, this term gives rise to density inhomo-
δρ
geneities ∼ 10−5 . This model is analogous to the Shafi–Vilenkin model, but it is much
ρ
simpler, and it is also shares none of the latter’s problems with baryogenesis. Likewise,
the extremely small coupling constant λ1 ∼ 10−14 once again need not be introduced
beforehand — the constant λ2 ∼ 10−6 is sufficient, which seems more natural, inasmuch
as similar constants do appear in such popular models as the Glashow–Weinberg–Salam
theory.

9.3 Chaotic inflation in supergravity

There are currently several different models that describe chaotic inflation in the context
of supergravity [273, 274, 282]. Here we examine one of these that seems to us to be
particularly simple, a model related to SU(n, 1) supergravity [283], several versions of
which arise in the low-energy limit of superstring theory [17].
One of the major problems that comes up in constructing realistic models based on
supergravity theory is how to make the effective potential V(z) vanish at its minimum
z0 . As a first step toward such a theory, one can attempt to find a general form of the
function G(z, z ∗ ) for which the potential V(z, z ∗ ) in (8.4.2) is identically zero. This occurs
when [284]
3
G(z, z ∗ ) = − ln(g(z) + g ∗ (z))2 , (9.3.1)
2
where g(z) is some arbitrary function. In that case, the Lagrangian is

∂µ g ∂ µ g
L = Gzz ∗ ∂µ z ∂ µ z = 3 . (9.3.2)
(g + g ∗ )2

All such theories with different g(z) are equivalent to one another after the transformation
g(z) → z. The Lagrangian
∂µ z ∂ µ z
L=3 (9.3.3)
(z + z ∗ )2
is invariant under the group of SU(1, 1) transformations

αz + iβ
z→ (9.3.4)
iγ z + δ

with real parameters α, β, γ, δ such that α δ + β γ = 1 [284]. Such theories have been
called SU(1, 1) supergravity for that reason.
One possible generalization of the function G(z, z ∗ ) of (9.3.1) that leads to a potential
V(z, z ∗ , ϕ, ϕ∗) ≥ 0, where ϕ is the scalar (inflation) field responsible for inflation, is
3
G=− ln(z + z ∗ + h(ϕ, ϕ∗ ))2 + g(ϕ, ϕ∗ ) , (9.3.5)
2
 185

where h and g are arbitrary real-valued functions of ϕ and ϕ∗ . In the theory (9.3.5),
2
1 g |gϕ |
V= e , (9.3.6)
|z + z ∗ |2 Gϕϕ∗
where Gϕϕ∗ = gϕϕ∗ + Gz hϕϕ∗ ≥ 0 if the kinetic term for the field ϕ has the correct
(positive) sign.
Cast in terms of the variables z and ϕ, the theory (9.3.5) looks rather complicated,
but it can be simplified considerably by diagonalizing the kinetic part of the Lagrangian,
reducing it to the form [285]
1 3 2
Lkin = ∂µ ζ ∂ µ ζ + e 3 ζ I2µ + Gϕϕ∗ ∂µ ϕ∗ ∂ µ ϕ , (9.3.7)
12 4
where
3
ζ = − ln(z + z ∗ + h(ϕ, ϕ∗ ))2 ,
2
Iµ = i [∂µ (z − z ∗ ) + hϕ ∂µ ϕ − hϕ∗ ∂µ ϕ∗ ] ,
Gϕϕ∗ = gϕϕ∗ + Gz hϕϕ∗ = gϕϕ∗ − 3 eζ/3 hϕϕ∗ . (9.3.8)

In terms of ζ, the potential becomes

|gϕ |2
V = eζ+g . (9.3.9)
Gϕϕ∗

As the simplest realization of the chaotic inflation scenario in this model [274], one can
consider the theory (9.3.5) with

g(ϕ, ϕ∗) = (ϕ − ϕ∗ )2 + ln |f (ϕ)|2 , (9.3.10)

while h(ϕ, ϕ∗ ) satisfies

hϕϕ∗ = (2 a)−1 gϕϕ∗ = −a−1 , (9.3.11)
where a is a positive constant. Then

∂V a − eζ/3
Vζ ≡ =V , (9.3.12)
∂ζ 3 ζ/3
a− e
2
that is, Vζ = 0 when eζ/3 = a. Notice that at the extremum of V (i.e., at Vζ = 0), the
field ϕ has the canonical kinetic term
1
Gϕϕ∗ = − gϕϕ∗ = 1 , (9.3.13)
2
and
2
Vζζ ∗ = V>0. (9.3.14)
3
 186

This means that the potential V(ϕ, ζ) has a hollow located at ζ = 3 ln a, −∞ < ϕ < ∞.
At the bottom of the hollow, the potential V(ϕ, ζ) is
2
V(ϕ) = a3 eg |gϕ |2 = a3 e−4 η |fϕ + 4 i η|2 , (9.3.15)
where ϕ = ξ + i η. Equation (9.3.15) implies that for all real ϕ,
V = a3 |fϕ |2 . (9.3.16)
This resembles the expression for the effective potential in a globally supersymmetric
theory with the superpotential f (ϕ). Inflation takes place in this theory for a broad
class of superpotentials, such as those with f (ϕ) ∼ ϕn , n > 1. A complete description
of inflation in this theory is quite complicated, particularly on account of the presence
of non-minimal kinetic energy terms in (9.3.7). The third term in (9.3.7), for example,
leads to an extra term ∼ a−1 eζ/3 |∂µ ϕ|2 in Vζ (9.3.12). Fortunately, |∂µ ϕ|2 ≪ V during
inflation, and the corresponding correction turns out to be negligible.
In order to study the evolution of the universe in this model, we assume that ϕ is
MP
originally a fairly large field, |ϕ| ≫ 1 (or |ϕ| ≫ √ in conventional units). Then
8π
both the curvature Vηη ∼ a3 |f |2 and the curvature Vζζ ∼ a3 |fϕ |2 are much greater than
the curvature Vξξ , which in the theory under consideration is of order a3 |fϕ |2 ϕ−2 . If
2a
ζ 6= 3 ln a (ζ > 3 ln ) from the outset and η 6= 0 (|η| < ∼ 1), then the fields ζ and
3
ϕ quickly roll down to the bottom of the hollow, where ζ = 3 ln a and η = 0, and the
effective potential is given by (9.3.16). The field ϕ then has the usual kinetic energy term
(9.3.13), and for f = µ3 ϕn ,
V(ϕ) = n2 a3 µ6 ϕ2 n−2 . (9.3.17)
In particular, if f = µ3 ϕ3 ,
V(ϕ) = 9 a3 µ6 ϕ4 . (9.3.18)
The universe undergoes inflation as the field ϕ rolls down from ϕ ≫ 1 to ϕ < ∼ 1. The
δρ
density inhomogeneities that are produced in the theory (9.3.18) are of order ∼ 10−5
√ ρ
when α µ ∼ 10−2 –10−3 . There is thus no need to introduce any anomalously small
coupling constants like λ ∼ 10−14 ; in this scenario, the combination a3 µ6 takes on that
7
role. A typical inflation factor for the universe in this model is of order 1010 . The process
whereby the universe is reheated depends on the manner in which the field ϕ interacts
8
with the matter fields. As a rule, reheating to a temperature TR > ∼ 10 GeV is readily
accomplished [286], enabling baryon asymmetry production by the mechanisms described
in Chapter 7.

9.4 The modified Starobinsky model and the combined

scenario
In all of the models discussed thus far, inflation is driven by an elementary scalar field
ϕ. However, the role of this field can also be played by a condensate of fermions hψ̄ ψi
 187

or vector particles hGaµν Gaµν i, or simply by the curvature scalar R itself. The latter
possibility provided the basis for the Starobinsky model [52], which could be considered
the first version of the inflationary universe scenario, and which predated even the model
of Guth. In its original form, this model was based on the observation of Dowker and
Critchley [106] that when the conformal anomaly of the energy-momentum tensor is taken
into account, de Sitter space with energy density approaching the Planck density turns
out to be a self-consistent solution of the Einstein equations with quantum corrections.
Starobinsky showed that the corresponding solution is unstable; the curvature scalar
starts to decrease slowly at some moment, and this decrease accelerates. Finally, after
the oscillatory stage, the universe is reheated, and is then described by the standard hot
universe model.
The formal description of the decay of the initial de Sitter space in the Starobinsky
model bears a close resemblance to the theory of the decay of the unstable state ϕ = 0
in the new inflationary universe scenario. When this model was proposed, it elicited an
enormous amount of interest from cosmologists [287]. But the origin of the unstable de
Sitter state in the Starobinsky model remained somewhat enigmatic — the conventional
wisdom was that either such a state came into being as a result of an asymmetric collapse
of a previously existing universe [288], or that the universe appeared “from nothing” in an
unstable quasivacuum state [289, 290]. These suggestions seemed rather more complicated
than the principles underlying the new inflationary universe scenario. Furthermore, the
original Starobinsky model, like the first versions of the new inflationary universe scenario,
δρ
leaves us with density inhomogeneities after inflation that are too large [107], and it
ρ
fails to provide a solution to the primordial monopole problem.
Subsequently, however, it proved to be possible to modify this model and imple-
ment it in a manner that was similar in spirit to the chaotic inflation scenario [108–110].
The crux of this modification entailed replacing the study of one-loop corrections to the
energy-momentum tensor Tνµ with an examination of a gravitational theory in which terms
R
quadratic in the curvature tensor Rµναβ are added to the Einstein Lagrangian .
16 π G
In general, this is far from an innocuous procedure, inasmuch as metric perturbations
then turn out to be described by fourth-order equations, and this frequently leads to
particles having imaginary mass (tachyons) or negative energy (indefinite metric) [291].
R2 M2P
Fortunately, these problems do not appear if just one term is added, with M2 ≪
96 π 2 M2
M2P . When the sign in front of R2 is correctly chosen, this term leads to the emergence
of a scalar excitation (scalaron) corresponding to a particle with positive energy and
mass M2 > 0. Taking the term ∼ R2 into consideration, the Einstein equations are then
modified. Specifically, in a flat Friedmann space (k = 0), Eq. (1.7.12) for the universe
filled by a uniform field ϕ is replaced by
 !2 
2
8π 1 2 H Ḧ Ḣ
 
H2 = 2
ϕ̇ + V(ϕ) − 2 Ḣ + 2 −  . (9.4.1)
3 MP 2 M H H
 188

Let us first neglect the contribution to (9.4.1) from the field ϕ, and consider solutions of
the modified Einstein equations in the absence of matter fields. Equation (9.4.1) will then
admit of a solution satisfying the conditions |Ḣ| ≪ H2 , |Ḧ| ≪ |Ḣ H|, or in other words, a
solution that describes an inflationary universe with a slowly changing parameter H [108,
109]:
1 2
H = M (t1 − t) , (9.4.2)
6
M2
!
2
a(t) = a0 exp (t1 − t) . (9.4.3)
12
These conditions prevail until H becomes smaller than M; after that, the stage of inflation
1
ends, and H starts to oscillate about some mean value H0 (t) ∼ . The universe then
t
heats up, and it can subsequently be described by the familiar hot universe theory.
Strictly speaking, Eqs. (9.4.2) and (9.4.3) are only applicable if R2 , Rµν Rµν ≪ M4P .
Moreover, depending on the initial value of H, inflation may start much later than the
Planck time, at which R2 and Rµν Rµν become of the same order as M4P . We thus arrive
once again at the problem of the evolution of a universe in which inflation takes place only
in regions with suitable initial conditions (in no way associated with high-temperature
phase transitions). In other words, we again wind up with the chaotic inflation scenario,
where the curvature scalar R (equal to 12 H2 at the time of inflation) takes on the role
of the scalar inflaton field. In the more general case in which scalar fields ϕ (9.4.1) are
also present, several different stages of inflation are possible, where the dominant effects
are either associated with the scalar fields or they are purely gravitational, as described
above [110].
The succession of different stages is governed by the relationship between √the mass
M of the scalaron and the mass m of the scalar field ϕ when ϕ ∼ MP (m ∼ λ MP in
λ
the ϕ4 theory). When m ≫ M, the stage in which the field ϕ dominates rapidly comes
4
to an end, and the next stage of inflation is associated with purely gravitational effects.
δρ
During that stage, as usual, density inhomogeneities are produced which on a galactic
ρ
scale are given to order of magnitude by [107, 221]
δρ M
∼ 103 , (9.4.4)
ρ MP
δρ
or in other words, ∼ 10−5 when
ρ
M ∼ 1011 GeV . (9.4.5)
Reheating of the universe in this instance is also driven by purely gravitational effects [52,
134]. According to (7.9.17),
s
q M3
TR ∼ 10 −1
Γ MP ∼ 10 −1
∼ 106 GeV . (9.4.6)
MP
 189

6
To account for baryogenesis at a temperature T < ∼ TR ∼ 10 GeV as in the case of
the Shafi–Vilenkin model and a number of other models based on supergravity, it is
necessary to invoke the nonstandard mechanisms described in Section 7.10. Note, however,
R2 M2P
that terms like , if they do appear in elementary particle theories or superstring
96 π 2 M2
theories, will do so, as a rule, with M ∼ MP rather than M ∼ 10−8 MP . It therefore seems
more likely that realistic theories will yield m ≪ M. The modified Starobinsky model
may then turn out to be responsible for the description of the earliest stages of inflation,
while the formation of the observable structure of the universe and its reheating take
place during the stage when the scalar field ϕ dominates. Reference 110 describes a more
detailed investigation of the combined model (9.4.1), effects related to the scalar field ϕ,
and effects associated with quadratic corrections to the Einstein Lagrangian.

9.5 Inflation in Kaluza–Klein and superstring theories

It was noted in Chapter 1 that our fondest hopes for constructing a unified theory of all
fundamental interactions have been tied in recent years to Kaluza–Klein and superstring
theories. One feature common to both of these theories is the proposition that original
space-time has a dimensionality d ≫ 4. Theories with d = 10 [17], d = 11 [16], d = 26
[94], and even d = 506 [95, 96] have all been entertained. The assumption is that d − 4
dimensions are compactified, and that space takes on dimensions of order M−1 P in the
corresponding directions, so that we are actually able to move only in the one remaining
time and three remaining space directions. It is usually assumed that the compactified
directions are spatial, but the possibility of compactifying multidimensional time has
also aroused some interest [292, 293]. The properties of a compactified space, in the
final analysis, determine the basic properties of the elementary particle theory that it
engenders.
Unfortunately, neither specific elementary particle models based on Kaluza–Klein and
superstring theories nor associated cosmological models have yet come close to fruition.
Nevertheless, it does make sense to examine the results that have been obtained in this
field.
One of the most interesting and detailed models of inflation based on Kaluza–Klein
models is that of Shafi and Wetterich [237]. The basis for this model is the Einstein action
with corrections that are quadratic in the d-dimensional curvature:
1 √
Z
S = − dd x gd
VD
n o
× α R̂2 + β R̂µ̂ν̂ R̂µ̂ν̂ + γ R̂µ̂ν̂ σ̂λ̂ R̂µ̂ν̂ σ̂λ̂ + δ · R̂ + ε . (9.5.1)

Here µ̂, ν̂, . . . = 0, 1, 2, . . . , d − 1; R̂µ̂ν̂ σ̂λ̂ is the curvature tensor in d-dimensional space;
VD is a volume in D-dimensional compactified space, D = d − 4; and α, β, and γ are
 190

dimensionless parameters. With

ζ = D (D − 1) α + (D − 1) β + 2 γ > 0 , (9.5.2)
δ > 0, (9.5.3)
1 2
ε = δ D (D − 1) ζ −1 , (9.5.4)
4
the equations for the d-dimensional metric have a solution of the form M4 × SD , where
M4 is a Minkowski space, and SD is a sphere of radius
2ζ
L20 = . (9.5.5)
δ
For
χ = (D − 1) β + 2 γ > 0 , (9.5.6)
the effective gravitational constant describing the interaction at large distances in M4
space is positive:
χ
G−1 = M2P = 16 π δ . (9.5.7)
ζ
A question remains concerning the stability of the solution M4 ×SD , but it has at least been
proven that with certain constraints on the parameters of the theory, compactification is
stable against variations of the radius of the sphere SD [294].
To describe cosmological evolution in this model, it is convenient to introduce the
four-dimensional scalar field
L(x)
ϕ(x) = ln . (9.5.8)
L0
After a suitable change of scale of the metric gµ̂ν̂ (x), the effective action in four-dimensional
space can be expressed as
√
Z
S = − d4 x g4
M2P R

× + exp D ϕ · (α R2 + β Rµν Rµν + γ Rµνσλ Rµνσλ )
16 π
1 2 1
− f (ϕ) ∂µ ϕ ∂ µ ϕ − fR (ϕ) · R ∂µ ϕ ∂ µ ϕ
2 2 
− h̃(ϕ) ∂µ ϕ ∂ µ R + V(ϕ) + ∆Lkin . (9.5.9)

Here µ, ν, . . . = 0, 1, 2, 3, and ∆Lkin takes in terms that comprise many derivatives of the
field ϕ, like ∂µ ∂ν ϕ · ∂ µ ∂ ν ϕ, etc. The potential V(ϕ) takes the form
!2 !2
M2P D (D − 1) −D ϕ 1 − e−2 ϕ
V(ϕ) = e , (9.5.10)
16 π 4ζ 1 − σ e−2 ϕ
 191

χ
where σ = − 1. The functions f 2 (ϕ), fR (ϕ), and h̃(ϕ) in (9.5.9) depend on α, β, γ, and
ζ
D. From (9.5.10), it follows that V(ϕ) ≥ 0; V(ϕ) tends to zero only when ϕ = 0 — that
is, when L(x) = L0 . When Rµνσλ 6= 0, however, the term
eD ϕ Kϕ = eD ϕ (α R2 + β Rµν Rµν + γ Rµνσλ Rµνσλ ) (9.5.11)

also contributes to the equation of motion of the field ϕ, and the function
W(ϕ) = V(ϕ) + eD ϕ Kϕ (9.5.12)
then plays the role of the potential energy of the field ϕ.
On the other hand, it is readily demonstrated that at the stage of inflation, the term
(9.5.11) makes a contribution ∼ eD ϕ H2 Ḣ to the Einstein equations in four-dimensional
space-time, and with H = const, this contribution can be neglected. In that approxi-
mation, then, the rate of inflation of the universe does not depend on the fact that the
additional term (9.5.11) is present, and is governed solely by the potential V(ϕ),
8π
H2 = V(ϕ) , (9.5.13)
3 M2P
while at the same time, the evolution of the field ϕ depends on the form of the potential
W(ϕ) (9.5.12):
∂W ∂V
3 H h2 (ϕ) ϕ̇ = − =− − D eD ϕ Kϕ (H(ϕ)) . (9.5.14)
∂ϕ ∂ϕ
The function h2 (ϕ) appears in (9.5.14) by virtue of the non-minimal nature of the kinetic
energy terms pertaining to the field ϕ in (9.5.9). This function varies slowly as ϕ changes,
∂W
and goes to a constant as ϕ → ∞. The function at large ϕ behaves as follows:
∂ϕ
!2
∂W M2P D 2 (D − 1)
lim = (µ − 1) e−D ϕ , (9.5.15)
ϕ→∞ ∂ϕ 16 π 4ζ
where
D−4
µ−1 = [3 (D − 1) β + 2 (D + 3) γ] . (9.5.16)
12 ζ
For µ > 1, the potential W(ϕ) approaches some constant from below, with the correspond-
ing difference becoming exponentially small. This implies that the field ϕ rolls down to
the minimum of W(ϕ) at ϕ = 0 exponentially slowly. On the other hand, the potential
V(ϕ), which determines the rate of expansion of the universe, is also exponentially small
at large ϕ. Nonetheless, with an initial value of ϕ > ∼ O(1) and a reasonable choice of
the constants α, β, and γ, it is possible simultaneously to obtain both a high degree of
inflation and small density perturbations. In particular, the duration of the inflationary
stage in this model is given approximately [237] by
2 K∞
∆t ∼ H−1 ϕ, (9.5.17)
µ−1
 192

where K∞ is defined by
M2P D
lim h2 (ϕ) = h2∞ = K∞ . (9.5.18)
ϕ→∞ 16 π 4 ζ
The quantity K∞ may be expressed in terms of α, β, and γ, and is usually of order unity.
It can easily be shown that by choosing α, β, γ ∼ 1, which is quite natural, one can obtain
∆t ∼ 70 H−1 , assuming an initial value ϕ ∼ 3 [237].
δρ
In this model, the quantity is given by an expression like (7.5.21), the only difference
ρ
being that ϕ̇ h(ϕ) appears instead of ϕ̇:
H2 H2 ∆t
1/2
δρ 2H πϕ H ∆t

∼ 0.2 ∼ 0.2 ∼ . (9.5.19)
ρ ϕ̇ h(ϕ) h∞ ϕ MP D K∞ ϕ
At the epoch of interest, H ∆t ∼ 70, and ϕ ∼ 3 so
δρ H
∼C , (9.5.20)
ρ MP
δρ
where C = O(1). In particular, ∼ 10−5 when
ρ
!1/2
H 1 D (D − 1) D
 
∼ exp − ϕ ∼ 10−5 . (9.5.21)
MP 8 6πζ 2
For ϕ ∼ 3, (9.5.20) is satisfied in theories with d = D + 4 = O(10). One interesting
feature of the perturbation spectrum obtained under these circumstances is its decline at
large ϕ — that is, at long wavelengths. This is related to the fact that the behavior of the
field and the rate of expansion of the universe in this model are determined by the two
different functions W(ϕ) and V(ϕ), respectively, rather than the single function V(ϕ).
The Shafi–Wetterich model is also interesting in that the curvature of the effective
potential W(ϕ) is quite large for ϕ ≪ 1. After inflation, ϕ oscillates in the vicinity of
ϕ = 0 at close to the Planck frequency, and the universe is reheated very rapidly and
efficiently. In this model, the temperature can reach TR ∼ 1017 GeV after reheating [295].
The main problem here is related to the initial conditions required for inflation. Indeed,
it would be unnatural to assume, within the framework of Kaluza–Klein theories, that
three-dimensional space has been infinite from the very beginning, since that would mean
that the distinction between compactified and non-compactified dimensions would have
to have been inserted into the theory from the outset, rather than arising spontaneously.
It would be more natural to suppose that the universe has been compact since its birth,
but that it has expanded at different rates in different directions: in three dimensions,
it has grown exponentially, while in d − 4 dimensions, it has gradually acquired a size of
approximately L0 ∼ M−1 P (9.5.5), (9.5.7). To phrase it differently, we are dealing with a
compact (closed, for example) universe governed by an expansion law that is asymmetric
in different directions.
 193

In Chapter 1, it was noted that a closed universe has a typical total lifetime of order
M−1
P , and the only thing that can rescue it from collapse is inflation that begins immedi-
ately after it has emerged from a state of Planck energy density. In the Shafi–Wetterich
model, however, inflation ought to begin when ϕ > ∼ 3, H <
−5
∼ 10 MP (9.5.21), that is, when
V(ϕ) ≪ M4P . In that event, inflation cannot save the universe from a premature death. In
order to circumvent this problem, it was suggested in [237] that the entire universe came
into being as a result of a quantum jump from the space-time foam (from “nothing”)
into a state with ϕ > ∼ 3, H <
−5
∼ 10 MP , a possibility we shall discuss in the next chapter.
Unfortunately, however, estimates of the !probability for such processes [296] lead to an
M4
expression of the form P ∼ exp − P , giving P ∼ exp(−1010 ) in the present case.
V(ϕ)
Thus, the outlook for a natural implementation of the inflationary universe scenario in
the context of the Shafi–Wetterich model is not very good. In fact, we are the victims
here of difficulties of the same type as those that prevent a successful implementation of
the new inflationary universe scenario.
It might be hoped that these problems will all go away when we make the transition
to a superstring theory. Such theories present several different candidates for the inflaton
field responsible for the inflation of the universe — it may be some combination of the
dilaton field that appears in superstring theory and the logarithm of the compactification
radius. Regrettably, our current understanding of the phenomenological and cosmological
aspects of superstring theory are still not entirely satisfactory. Existing models of inflation
based on superstring theories [297] rest on various assumptions about the structure of
these theories, and these assumptions are not always well-founded. But it is the initial
conditions, as before, that are the main problem. Our view is that the initial conditions
prerequisite to the onset of inflation in most of the models based on superstring theories
that have been proposed thus far are unnatural.
Does this mean that we are headed down the wrong road? At the moment, that is a
very difficult question to answer. It is quite possible that with the future development of
superstring theory, the inflationary universe scenario will be implemented in the context
of the latter in some nontrivial way (see, for example, Ref. 353). On the other hand,
one should recall that over the past decade, three palace revolts have taken place in
the land of elementary particles. In place of grand unified theories came theories based
on supergravity, followed by Kaluza–Klein theories, and finally the presently reigning
superstring theory. The inflationary universe scenario can be successfully implemented in
some of these theories; in some, this has not yet been accomplished, but there are no “no-
go” theorems that say it is impossible. In our opinion, what we have encountered here is
a somewhat nonstandard aspect of a standard situation. A theory should be constructed
in such a way that it describes experimental data, but this cannot always be done, and
the theory must then be changed. Until recently, however, cosmological data have not
been counted among the most important experimental facts. This situation has now
been radically altered, and it might just be that models in which inflation of the universe
cannot be implemented in a natural way will be rejected as being inconsistent with the
experimental data (if, of course, we find no alternative solution for all of the cosmological
 194

problems outlined in Chapter 1 that is not based on the inflationary universe scenario). In
analyzing the current state of affairs in this field, it must also be borne in mind that our
understanding of the inflationary universe scenario, and in particular the most important
question of initial conditions, is far from complete. In recent years, our conception of the
initial-condition problem in cosmology and our ideas about the global structure of the
inflationary universe have undergone a significant change. Progress in this field depends
primarily on the development of quantum cosmology, the topic to which we now turn.
 10
Inflation and Quantum Cosmology
If a man will begin with certainties, he shall end
in doubts; but if he will be content to begin
with doubts he shall end in certainties.

Francis Bacon (1561–1626)

The Advancement of Learning, Book V

10.1 The wave function of the universe

Quantum cosmology is conceptually one of the most difficult branches of theoretical
physics. This is due not just to such difficulties as the ultraviolet divergences encoun-
tered in the quantum theory of gravitation, but also in large measure to the fact that
the very formulation of the problems studied in quantum cosmology is not at all a trivial
matter. The results of research often appear paradoxical, and it requires an especially
open mind not to dismiss them outright.
The foundations of quantum cosmology were laid at the end of the 1960’s by Wheeler
and DeWitt [298, 299]. But prior to the advent of the inflationary universe scenario, a
description of the universe as a whole within the framework of quantum mechanics seemed
to most scientists to be an unnecessary luxury. When one describes macroscopic objects
using quantum mechanics, the results are usually the same as those given by classical
mechanics. If the universe is indeed the largest macroscopic entity that exists, then why
bother to describe it using quantum theory?
In the standard hot universe theory, this was a perfectly legitimate question, since ac-
cording to that theory the observable part of the universe resulted from the expansion of a
region containing a total of perhaps 1087 elementary particles. But in the inflationary uni-
verse scenario, the entire observable part of the universe (and possibly the entire universe
−1
itself) was formed by virtue of the rapid expansion of a region of size l <∼ MP ∼ 10
−33

cm containing perhaps not a single elementary particle! Quantum effects could thus have
played a pivotal role in events during the earliest stages of expansion of the universe.
Until recently, the fundamental working tool in quantum cosmology has been the
Wheeler–DeWitt equation for the wave function of the universe Ψ(hij , ϕ), where hij is
the three-dimensional spatial metric, and ϕ is the matter field. The Wheeler–DeWitt
 196

equation is essentially the Schrödinger equation for the wave function in the stationary
∂Ψ
state given by = 0 (see below). It describes the behavior of the quantity Ψ in so-
∂t
called superspace — the space of all three-dimensional metrics hij (not to be confused
with the superspace used to describe supersymmetric theories!). A detailed exposition
of the corresponding theory may be found in [298–301]. But the most interesting results
in this sphere were obtained using a simplified approach in which only a portion of the
full superspace, known as a minisuperspace, was considered, giving a description of a
homogeneous Friedmann universe; the scale factor of the universe a took up the role
of all the quantities hij . In this section, we will therefore illustrate the basic problems
relating to the calculation and interpretation of the wave function of the universe with an
example of the minisuperspace approach. In subsequent sections, we will discuss the limits
of applicability of this approach, the results obtained via recently developed stochastic
methods for describing an inflationary universe [134, 135, 57, 132, 133], and a number of
other questions with a bearing on quantum cosmology.
Let us consider, then, the theory of a scalar field ϕ with the Lagrangian
R M2P 1
L(gµν , ϕ) = − + ∂µ ϕ ∂ µ ϕ − V(ϕ) (10.1.1)
16 π 2
in a closed Friedmann universe whose metric can be represented in the form

ds2 = N2 (t) dt2 − a2 (t) dΩ23 , (10.1.2)

where N(t) is an auxiliary function that defines the scale on which the time t is measured,
and dΩ23 = dχ2 + sin2 χ (dθ2 + sin2 θ dϕ2 ) is the element of length on a three-dimensional
sphere of unit radius. To obtain an effective Lagrangian that depends on a(t) and ϕ(t),
one must integrate over angular variables in the expression for the action S(g, ϕ), which
√
with the factor g taken into account gives 2 π 2 a3 . Then, making use of the fact that
the universe is closed (has no boundaries), one obtains the action in a form that depends
only on a and ȧ, but not on ä:
3 M2P π ȧ2 a ϕ̇2
! !
L(a, ϕ) = − − N a + 2 π 2 a3 N − V(ϕ) . (10.1.3)
4 N 2 N2
The canonical momenta are
∂L 2 π 2 a3
πϕ = = ϕ̇ , (10.1.4)
∂ ϕ̇ N
∂L 3 M2P π
πa = =− ȧ a , (10.1.5)
∂ ȧ 2N
∂L
πN = =0, (10.1.6)
∂ Ṅ
and the Hamiltonian is

H = πϕ ϕ̇ + πa ȧ − L(a, ϕ)
 197

πa2 3 π M2P 2 πϕ2

! !
N N
= − 2
+ a + + 2 π 2 a4 V(ϕ)
a 3 π MP 4 a 4 π 2 a2
= Ha + Hϕ . (10.1.7)

Here Ha and Hϕ are the effective Hamiltonians of the scale factor a and scalar field ϕ in
the Friedmann universe. The equation relating the canonical variables πa , πϕ , a, and ϕ
follows from (10.1.6):

πa2 3 π M2P
!
∂H H 1
0= = = − + a
∂N N a 3 π M2P 4
πϕ2
!
1 2 4
+ + 2 π a V(ϕ) . (10.1.8)
a 4 π 2 a2

Upon quantization, Eq. (10.1.8) gives the relation that governs the wave function of the
universe:
∂Ψ(a, ϕ)
i = HΨ = 0 . (10.1.9)
∂t
In the usual fashion, the canonical variables are replaced with the operators
1 ∂
ϕ → ϕ, πϕ →
i ∂ϕ
1 ∂
a → a, πa → , (10.1.10)
i ∂a
and Eq. (10.1.9) takes the form

∂2 3πM2P 2 ∂2
!
1 1
− + a + − 2π 2 a4 V(ϕ) Ψ(a, ϕ) = 0 . (10.1.11)
3πM2P ∂a2 4 4π 2 a2 ∂ϕ2

This then is the Wheeler–DeWitt equation in minisuperspace.

Strictly speaking, it should be pointed out that ambiguities relating to the commuta-
tion properties of a and πa can arise in the derivation of Eq. (10.1.11). Instead of the term
∂2 1 ∂ p ∂
− 2 in (10.1.11), one sometimes writes − p a , where the parameter p can take
∂a a ∂a ∂a
on various values. In the semiclassical approximation, which will be of particular interest
to us later on, the actual value of this parameter is unimportant, and in particular, one
can take p = 0 and seek a solution of (10.1.11).
Clearly, however, Eq. (10.1.11) has many different solutions, and one of the most
fundamental questions facing us is which of these solutions actually describes our universe.
Before launching into a discussion of this question, we make several remarks of a general
nature that bear upon the interpretation of the wave function of the universe.
First of all, we point out that the wave function of the universe depends on the scale
factor a but, according to (10.1.9), it is time-independent. How can this be reconciled
with the fact that our observable universe does depend on time?
 198

Here we encounter one of the principal paradoxes of quantum cosmology, a proper

understanding of which is exceedingly important. The universe as a whole does not de-
pend on time because the very concept of such a change presumes the existence of some
immutable reference that does not appertain to the universe, but relative to which the
latter evolves. If by “the universe” we mean “everything,” then there remains no external
observer according to whose clocks the universe could develop. But in actuality, we are
not asking why the universe is developing, we are inquiring as to why we perceive it to
be developing. We have thereby separated the universe into two parts: a macroscopic
observer with clocks, and all the rest. The latter can perfectly well evolve in time (ac-
cording to the clocks of the observer), despite the fact that the wave function of the entire
universe is time-independent [299].
In other words, we arrive at our customary picture of a world that evolves in time
only after the universe has been divided into two macroscopic parts, each of which devel-
ops quasiclassically. The situation that ensues is reminiscent of the theory of tunnelling
through a barrier: the wave function is defined inside the barrier, but it yields the prob-
ability amplitude of finding a particle propagating in real time only outside the barrier,
where classical motion is allowed. By analogy, the universe too exists in its own right,
in a certain sense, but one can only speak of its temporal existence in the context of the
quasiclassical evolution of the part that remains after a macroscopic observer with clocks
has emerged.
Thus, by the very fact of his existence, an observer somehow reduces the overall wave
function of the universe to that part which describes the world that is observable to him.
This is exactly the point of view espoused in the standard Copenhagen interpretation of
quantum mechanics — the observer becomes not just a passive viewer, but something
more like a participant in the creation of the universe [302].
The situation is somewhat different in the many-worlds interpretation of quantum
mechanics [303–309], which presently enjoys a sizable following among quantum cosmol-
ogists. In this interpretation, the wave function Ψ(hij , ϕ) simultaneously describes all
possible universes together with the observers (of all possible kinds) that inhabit them.
In performing a measurement, rather than reducing the wave function of all of these
universes to the wave function of one of them (or a part of one of them), an observer
simply refines the issue of who he is and in which of these universes he resides. The
same results are then obtained as in the standard approach, but without recourse to the
somewhat ill-founded assumption of the reduction of the wave function at the instant of
measurement.
We shall not engage here in a detailed discussion of the interpretation of quantum me-
chanics, a problem which becomes particularly acute in the context of quantum cosmology
[302, 309]; instead, we return to our discussion of the evolution of the universe.
Another manifestation of the fact that the universe as a whole does not change in
time is that the wave function Ψ(a, ϕ) depends only on the quantities a and ϕ, and
not on whether the universe is contracting or expanding. One might interpret this to
mean that at the point of maximum expansion of a closed universe, the arrow of time is
somehow reversed, the total entropy of the universe begins to decrease — and observers
 199

are rejuvenated [310]. However, to determine the direction of the arrow of time, one
must first divide the universe into two quasiclassical subsystems, one of which contains an
observer with clocks. In general, the wave function of each such subsystem will certainly
not be symmetric under a change in the sign of ȧ. After the division of the universe into
two semiclassical subsystems, one can make use of the usual classical description of the
universe, according to which the total entropy of the universe can only increase with time,
and there is no way in which the direction of the arrow of time can ever be reversed at
the instant of maximum expansion [311].
We have discussed all these problems in such detail here in order to demonstrate that
not just the solution but even the formulation of problems in the context of quantum
cosmology is a nontrivial matter. The question of whether entropy can decrease in a
contracting universe, whether the arrow of time can be reversed in a singularity or at
the point of maximum expansion of a closed universe, whether the universe can oscillate,
has thus far bothered many experts in quantum cosmology; see, for example, [312, 313].
Above, we enunciated our own viewpoint in this regard, but it should be understood that
the comprehensive investigation of these questions is only just beginning.
The Wheeler–DeWitt equation (10.1.11) has many different solutions, and it is very
difficult to ascertain which of them actually describes our universe. One of the most
interesting suggestions here was advanced by Hartle and Hawking [314], who proposed
that the universe possesses a ground state, or a state of least excitation, similar to the
vacuum state in quantum field theory in Minkowski space. By carrying out short-time
measurements in Minkowski space, one can see that the vacuum is not empty, but is
instead filled with virtual particles. Similarly, the universe that we observe might be
a virtual state (but with a very long lifetime, due to inflation), and the probability of
winding up in such a state might be determinable if the wave function of the ground state
of the universe were known. According to the Hartle and Hawking hypothesis, the wave
function Ψ(a, ϕ) of the ground state of the universe with scale factor a which is filled with
a homogeneous field ϕ is given in the semiclassical approximation by

Ψ(a, ϕ) ∼ N e−SE (a,ϕ) . (10.1.12)

Here N is a normalizing factor, and SE (a, ϕ) is the Euclidean action corresponding to

solutions of the equations of motion for a(ϕ(τ ), τ ) and ϕ(τ ) with boundary conditions
a(ϕ(0), 0) = a(ϕ), ϕ(0) = ϕ in space with a metric that has Euclidean signature.
The reason for choosing this particular solution of the Wheeler–DeWitt equation was
explained as follows. Consider the Green’s function of a particle that moves from the
point (0, t′) to (x, 0):
′
Ψn (x) Ψn (0) ei En t
X
hx, 0|0, t′i =
n
Z
= dx(t) exp{i S[x(t)]} , (10.1.13)

where Ψn (x) is a time-independent eigenfunction of the energy operator with eigenvalue

En ≥ 0. Let us now perform a rotation t → −i τ and take the limit as τ ′ → −∞. The
 200

only term that survives in the sum (10.1.13) is the one corresponding to the smallest
eigenvalue En (normalized to zero). This implies that
Z
Ψ0 (x) ∼ N dx exp{−SE [x(τ )]} . (10.1.14)

Hartle and Hawking have argued that the generalization of this result to quantum cos-
mology in the semiclassical approximation will yield (10.1.13). For a slowly varying field
ϕ (and this is precisely the most interesting case from the standpoint of implementing
the inflationary universe scenario), the solution of the Euclidean version of the Einstein
equations for a(ϕ, τ ) is

a(ϕ, τ ) ≈ H−1 (ϕ) cos[H(ϕ) τ ] ≡ a(ϕ) cos[H(ϕ) τ ] , (10.1.15)

v
u 8 π V(ϕ)
u
where H(ϕ) = t , and the corresponding Euclidean action is
3 M2P

3 M4P
SE (a, ϕ) = − , (10.1.16)
16 V(ϕ)

whereupon

3 M4P π M2P
! !
Ψ[a(ϕ), ϕ] ∼ N exp = N exp
16 V(ϕ) 2 H2(ϕ)
π M2P a2 (ϕ)
!
= N exp . (10.1.17)
2

Hence, it should follow that the probability of detecting a closed universe in a state with
field ϕ and scale factor a(ϕ) = H−1 (ϕ) is

3 M4P
!
2 2 2
P[a(ϕ), ϕ] ∼ N |Ψ[a(ϕ), ϕ)]| ∼ N exp
8 V(ϕ)
2 2 2
= N exp[π MP a (ϕ)] . (10.1.18)

If the ground state of the universe were a state with ϕ = ϕ0 and 0 < V(ϕ0 ) ≪ M4P , then
the normalization factor N2 that ensures a total probability of all realizations being unity
would have to be
3 M4P
!
2 2
N ∼ exp[−π MP a0 ] = exp − , (10.1.19)
8 V(ϕ0 )
where a0 = H−1 (ϕ0 ). From Eqs. (10.1.18) and (10.1.19), it follows that

3 M4P
" !#
1 1
P[a(ϕ), ϕ] ∼ exp − ) . (10.1.20)
8 V(ϕ) V(ϕ0
 201

To calculate the probability that a ≪ a0 = H−1 (ϕ0 ) or a ≫ a0 = H−1 (ϕ0 ) when ϕ = ϕ0 ,

we must venture outside the confines of the quasiclassical approximation (10.1.12) or solve
Eq. (10.1.11) directly in the WKB approximation. According to [314],
π 2 2
 
Ψ(a ≪ a0 ) ∼ exp MP (a − a20 ) , (10.1.21)
2
i H(ϕ0 ) M2P a3
" #
Ψ(a ≫ a0 ) ∼ exp
3
i H(ϕ0 ) M2P a3
" #
+ exp − . (10.1.22)
3
Unfortunately, the arguments used by Hartle and Hawking to justify (10.1.12) are far
from universally applicable. In fact, the Euclidean rotation alluded to above can be used
to eliminate all but the zeroth-order term from (10.1.13) only if En > 0 for all n > 0. While
the energy of excitations of the scalar field ϕ is positive, the energy of the scale factor a
is negative, so that these sum to zero; see (10.1.7) and (10.1.9). In such a situation, there
is no general prescription for extracting the ground state Ψ0 from the sum (10.1.13) by
rotation to Euclidean space. To investigate the properties of the field ϕ on scales much
smaller than the size of the closed universe, this is an unimportant issue, and we can
simply quantize the field ϕ against the background of the classical gravitational field and
perform the standard Euclidean rotation t → −i τ . This is exactly the reason why the
probability density function (10.1.20) is the same as the distribution (7.4.7), which was
derived using more conventional methods. On the other hand, in those situations where
the scale factor a itself must be quantized (for instance, in a description of the quantum
creation of the universe from a state with a = 0, i.e., from “nothing” [315–317, 289, 290,
318]), the corresponding problem becomes much more serious.
Fortunately, this can be avoided if the quantum properties of the field ϕ are unim-
portant for our purposes at the epoch of interest — for example, if ϕ is a classical slowly
varying field whose sole role is to produce a nonzero vacuum energy V(ϕ) (cosmological
term). One can then neglect quantum effects associated with the scalar field, and to
isolate the ground state Ψ(a, ϕ), corresponding to the lowest excitation state of the scale
factor a, one need only carry out the rotation t → +i τ in order to suppress the contribu-
tion to (10.1.13) from negative-energy excitations.1 This gives contribution to (10.1.13)
from negativee-energy excitations.1 This gives
3 M4P
!
SE (a,ϕ)
Ψ(a, ϕ) ∼ N e ∼ N exp − , (10.1.23)
16 V(ϕ)
and the probability that the universe appears in a state with field ϕ is
3 M4P
!
2 2
P[a(ϕ), ϕ)] ∼ |Ψ| ∼ N exp − . (10.1.24)
8 V(ϕ)
1
It should be borne in mind that there are no physical excitations of the gravitational field with negative
energy. Therefore, for a consistent quantization of the scale factor a one should introduce Faddeev–Popov
ghosts. However, as usual, ghosts do not contribute to Ψ(a, ϕ) in the semiclassical approximation.
 202

Equation (10.1.23) was first obtained using the method described above [319]; it was
later derived by Zeldovich and Starobinsky [320], Rubakov [321], and Vilenkin [322] using
a different method. For the reasons to be discussed soon, we will call (10.1.23) tunneling
wave function.
The obvious difference between Eqs. (10.1.24) and (10.1.18)–
(10.1.21) is in the sign of the argument of the exponential. This difference is extremely
important, since according to (10.1.18) and (10.1.20), the probability of detecting the
universe in a state with a large value of V(ϕ) is exponentially small. In contrast, Eq.
(10.1.24) tells us that the universe is most likely created in a state with V(ϕ) ∼ M4P . This
is consistent with our previous expectations, and leads to a natural implementation of the
chaotic inflation scenario [319].
In order to comprehend the physical meaning of the Hartle–Hawking wave function
(10.1.12), let us compare the solutions of Eqs. (10.1.21) and (10.1.22) with solutions
for the scalar field (1.1.3). One possible interpretation of the solution (10.1.21) is that
2 3
" #
i H(ϕ0 ) MP a
the wave function exp describes a universe with decreasing scale factor
3
a (compare with the wave function of #a particle with momentum p, ψ ∼ e−i p x ), while
i H(ϕ0 ) M2P a3
"
the wave function exp − corresponds to a universe with increasing scale
3
factor. Bearing in mind, then, that the corresponding motion takes place, according to
(10.1.11), in a theory for which the effective potential of the scale factor is
3 π M2P 2
V(a) = a − 2 π 2 a4 V(ϕ) , (10.1.25)
4
the interpretation of the solutions (10.1.21) and (10.1.22) becomes quite straightforward
(although different authors are still not in complete agreement on this point). The wave
function (10.1.22) describes a wave incident upon the barrier V(a) from the large-a side
and a wave reflected from the barrier; when a < H−1 (i.e., incidence below the barrier
height), the wave is exponentially damped in accordance with (10.1.21) (see Fig. 10.1).
The physical meaning of this solution is most easily grasped if one recalls that a closed
de Sitter space with V(ϕ0 ) > 0 first contracts and then expands: a(t) = H−1 cosh(H t).
The Hartle–Hawking wave function (10.1.21) accounts for the “broadening” of this quasi-
classical trajectory, and allows for the fact that at the quantum level, the scale factor can
become less than H−1 at the point of maximum contraction. The absence of exponential
suppression for a > H−1 (10.1.22) is related to the fact that values a > H−1 are classi-
cally allowed [314]. Observational cosmological data put the present-day energy density
of the vacuum V(ϕ0 ) at no more than 10−29 g/cm3 , which corresponds to H−1 > 28
∼ 10 cm.
The evolution of a de Sitter space whose minimal size exceeds 1028 cm has nothing in
common with the evolution of the universe in which we now live. Therefore, within the
scope of the foregoing interpretation, the Hartle–Hawking wave function does not give a
proper description of our universe in the minisuperspace approximation studied above.
We face a similar difficulty if we attempt (without justification) to use this wave function
instead of the function (10.1.23) to account for the very earliest stages in the evolution
 203

B
0 a0 a

Figure 10.1: The effective potential V(a) for the scale factor a of (10.1.25). This figure also
gives a somewhat tentative representation of the Hartle–Hawking wave function (10.1.22)
(curve A), and of the wave function (10.1.23), which describes the quantum birth of the
universe from the state a = 0 (curve B).

of the universe, since according to (10.1.18) and (10.1.20), the likelihood of a prolonged
inflationary stage would in that case be exponentially small.
One might suggest another possible interpretation of the Hartle–Hawking wave func-
tion, namely that its square gives the probability density function for an observer to detect
that he is in a universe of a given type not at the instant of its creation, but at the instant
of his first measurement, prior to which he cannot say anything about the evolution of
the universe.2 Such an interpretation may turn out to be eminently reasonable (and,
in the final analysis, independent of the choice of observer) if, as originally supposed by
Hartle and Hawking, a ground state actually exists for the system in question, so that the
probability distribution under consideration turns out to be stationary, like the vacuum
state or ground state of an equilibrium thermodynamic system. And in fact, as we have
already noted, the Hartle–Hawking wave function provides a good description of the qua-
sistationary distribution of the field ϕ in an intermediate metastable state (see (10.1.20)
and (7.4.7)).
On the other hand, a stationary distribution (10.1.20) of the field ϕ is only possible if
2 d2 V V(ϕ)
m = ≪ H2 ∼ in the vicinity of the absolute minimum of V(ϕ). Not a single
dϕ 2 M2P
realistic model of the inflationary universe satisfies this requirement. The only stationary
distribution of a (quasi)classical field ϕ in realistic models that we are presently aware
of (see the discussion of this question in Sections 7.4, 10.2, and 10.3) is the trivial delta-
function distribution, with the field totally concentrated at the minimum of V(ϕ). But
this is not at all the result that researchers in quantum cosmology are trying to obtain
2
From this standpoint, one might tentatively say that the wave function (10.1.23) is associated with the
creation of the universe, while (10.1.12) is associated with the creation of an observer.
 204

when they discuss the Hartle–Hawking wave function and assume that the appropriate
probability distribution is given by Eq. (10.1.20).
All these caveats notwithstanding, we would rather not draw any hasty conclusions,
which would be a particularly dangerous thing to do in a science whose ultimate founda-
tions have yet to be laid. The mathematical structure proposed by Hartle and Hawking is
quite elegant in and of itself, and quite possibly we may yet find a way to take advantage
of it. Our fundamental objection to the possibility of a stationary distribution of the field
ϕ in an inflationary universe is based upon a study of a (typical) situation in which the
field possesses one (or a few) absolutely stable vacuum states. But instances are known in
which theories are characterized by the value of some time-independent field, topological
invariant, or other parameter characterizing properties of the vacuum state which might
govern, say, the strength of CP violation, the energy of the vacuum, and so forth.
One such parameter is the angle θ which characterizes the vacuum properties in quan-
tum chromodynamics [183]. It is possible that the cosmological term and many coupling
constants in elementary particle theory [345, 346, 349] are also vacuum parameters of
this type. Their time-independence may be guaranteed by superselection rules of some
sort [346]. But in the context of the many-worlds interpretation of quantum mechanics,
the question of what the properties of the world (of the vacuum state) are in which the
observer finds himself at the instant of his first observation is a perfectly reasonable one.
The suggestion that the appropriate probability distribution will be given by the square
of the Hartle–Hawking wave function [346] seems to us worthy of serious consideration.
At the same time, the question of choosing between the Hartle–Hawking function and
the function (10.1.23) under these circumstances becomes especially important. As we
pointed out earlier, the Hartle–Hawking wave function actually gives the proper results
when one considers the (quasi)stationary distribution of a scalar field ϕ with positive
energy in a classical de Sitter background (see (10.1.20)). On the other hand, if the evolu-
tion of matter fields (and vacuum parameters) is insignificant, then the wave function may
possibly be determined by an expression like (10.1.23). We will return to the discussion
of this question in Section 10.7.
A possible interpretation (and an alternative derivation) of the wave function (10.1.23)
can also be obtained by studying tunneling through the barrier (10.1.25), but from the
direction of small a rather than large a [320–322]. Indeed, one can readily show that (with
π 2 2
 
ϕ ≈ const) a solution of Eq. (10.1.11) exists, which behaves as exp − MP a for
2
a < a0 = H−1 (ϕ) ≫ M−1
P

(cf. (10.1.21)), while for a ≫ H−1 (ϕ), it is represented by a wave

π M2P a20 i H M2P a3

!
∼ exp − −
2 3

emerging from the barrier and moving off toward large a; see Fig. 10.1. Damping of the
 205

wave as it emerges from the barrier is of order

π M2P a20 3 M4P
! !
exp − ∼ exp − ,
2 16 V(ϕ)
which exactly corresponds to Eq. (10.1.23) above. The wave function (10.1.23) thus
describes the process of quantum creation of a closed inflationary universe filled with a
homogeneous field ϕ due to tunneling from a state with scale factor a = 0, or in other
words, from “nothing” [319–322].
We now attempt to provide a plausible interpretation of this result, and to clarify the
reason why the probability of the quantum creation of a closed universe only becomes
large when V(ϕ) ∼ M4P . To this end we examine a closed de Sitter space with energy
density V(ϕ). Its volume at the epoch of maximum contraction (t = 0) is of order
H−3 (ϕ) ∼ M3P [V(ϕ)]−3/2 , and the total energy of the scalar field contained in de Sitter
M3
space at that instant is approximately E ∼ V(ϕ) H−3 (ϕ) ∼ q P . When V(ϕ) ∼ M4P ,
V(ϕ)
the total energy of the scalar field is E ∼ MP . By the uncertainty principle, one cannot rule
out the possibility of quantum fluctuations of energy E lasting for a time ∆t ∼ E−1 ∼ M−1 P .
However, within a time of this order (or slightly longer), de Sitter space of initial size
∼ H−1 ∼ M−1 P becomes exponentially large, and one can consider it to be an inflationary
universe emerging from “nothing” (or from the space-time foam). For small V(ϕ), the
probability that this process will come to pass should be extremely low, since as V(ϕ)
decreases, the minimum energy E of a scalar field in de Sitter space increases, rather than
falling off, and the typical lifetime of the corresponding quantum fluctuation becomes
much shorter than the Planck time.
It is important to note here that we are dealing with the creation of a compact universe
with no boundaries, so that no supplementary conditions such as ϕ = 0 are required at the
boundary of the bubble that is formed (compare this with the discussion of the “budding”
of the universe from Minkowski space in Section 10.3). If the effective potential V(ϕ) is
flat enough, so that the field ϕ rolls down to its minimum in a time much longer than
H−1 , then at the instant of its creation, the universe will “know nothing” of the location
of the minimum of V(ϕ) or how far the initial field ϕ is displaced from it. To a first
approximation, the probability of creation of the universe is given solely by the magnitude
of V(ϕ), in accordance with (10.1.24).
Generally speaking, the extent to which the probability of creation of a universe with
V(ϕ) ≪ M4P is suppressed may be lessened somewhat if particle creation at the time
of tunneling is taken into account [321]. Furthermore, exponential suppression may be
absent, by and large, during the creation of a compact (flat) universe with nontrivial
topology [320]. For us, the only important thing is that, as expected, there is no exponen-
tial suppression of the probability for creation of an inflationary universe with V(ϕ) ∼ M4P
— that is, from the present point of view, the initial conditions for implementation of the
chaotic inflation scenario are also found to be quite natural.
Note that the distinction between the creation of the universe from a singularity and
quantum creation from “nothing” at Planck density is rather tentative. In either case,
 206

one is dealing with the emergence of a region of classical space-time from the space-time
foam. The terminological distinction consists of the fact that by creation from “nothing,”
one usually means that a description of the evolution of the universe according to the
classical equations of motion begins only at large enough a. But due to large quantum
4
fluctuations of the metric at ρ > ∼ MP , it also turns out to be impossible to describe the
universe near the singularity (i.e., at small a). An important feature of either case is
that as a → 0, only quantum cosmology can provide a description of the evolution of the
universe, a circumstance that can lead to surprising consequences.
Consider, for example, a possible model for the evolution of a closed inflationary
universe. This model will be incomplete, and aspects of its interpretation will be open
to argument, but on the whole it furnishes a good illustration of some novel possibilities
being discussed within the context of quantum cosmology.
Thus, suppose that the universe was originally in a state a = 0. Quantum fluctuations
of the metric at that time were extremely large, and there were neither clocks nor rulers.
Any observations made by an imaginary observer at that epoch would have been uncorre-
lated with one another, and one could not even have said which of those observations came
first or last. The results of measurements could not be remembered, which implies that
with each new measurement the observer would effectively find himself in a completely
new space. If during one of these measurements he found himself inside a hot universe
that was not passing through an inflationary stage, then the characteristic lifetime of
such a universe would turn out to be of order M−1 P and its total energy E ∼ MP ; it would
therefore be essentially indistinguishable from a quantum fluctuation. But if the observer
detected that he was in an inflationary universe, he would then be able to make clocks
and rulers, and over an exponentially long period of time he could describe the evolution
of the universe with the aid of the classical Einstein equations. After a certain time, the
universe would be reheated, following which it would proceed through a state of maximum
expansion and begin to contract. When it had reached a state of Planck density (which
would occur when a ≫ M−1 P ), it would become impossible to use clocks and rulers and
thereby introduce any meaningful concept of time, entropy density, and so forth, due to
large quantum fluctuations of the metric. One could say that quantum fluctuations near
the singularity in effect erase from the memory of the universe any information about the
properties it had during its period of quasiclassical evolution. Consequently, after Planck
densities have been attained, subsequent observations again become disordered; it is even
dubious, strictly speaking, that one could say they are subsequent. At some point, the
observer detects that he is in an inflationary universe, and everything begins anew. Here
the parameters of the inflationary universe depend only on the value of the wave function
Ψ(a, ϕ), and not on its history, which has been “forgotten” during the passage through
the purgatory of Planck densities. We thus obtain a model for an oscillating universe
in which there is no increase of entropy during each successive cycle [298, 323]. Other
versions of this model exist, and are based on a hypothesized limiting density ρ ∼ M4P
4
[313] or gravitational confinement at ρ > ∼ MP [116].
The examples considered in this section indicate how interesting the investigation
of solutions of the Wheeler–DeWitt equation is, and by the same token, how difficult
 207

it is to choose and interpret a satisfactory solution. Studies of this question are only
just beginning [324]. Some of the problems encountered are related to the minisuperspace
approximation employed, and some to the fact that we wish to derive (or guess) the correct
solution to the full quantum mechanical problem without an adequate understanding of
the properties of the global structure of the inflationary universe at a more elementary
level. To fill this gap, it will prove very useful to study the properties of the inflationary
universe via the stochastic approach to inflation, which occupies an intermediate position
between the classical description of the inflationary universe and an approach based on
solving the Wheeler–DeWitt equation.

10.2 Quantum cosmology and the global structure of the

inflationary universe
One of the principal shortcomings of an approach based on minisuperspace is the initial
presumption of global homogeneity of the universe. The only explanation for the homo-
geneity of the universe which is known at present is based on the inflationary universe
scenario. However, as we showed in Section 1.8, effects related to long-wave fluctuations
of the scalar field prevent the geometry of an inflationary universe from having anything
in common with that of a homogeneous Friedmann space on much larger scales (at l > ∗
∼ l ).
Instead of a homogeneous universe that comes into being as a whole at some instant of
time t = 0, we must deal with a globally inhomogeneous self-regenerating inflationary
universe, whose evolution has no end and quite possibly may not have had a unique be-
ginning. Thus, many of the most important properties of the inflationary universe cannot
be understood or studied within the context of the minisuperspace approach.
In Chapter 1, we presented the simplest description of the mechanism of self-regeneration
of inflationary domains of the universe in the chaotic inflation scenario [57]. Below, we
engage in a more detailed investigation of this problem [132, 133].
One could carry out this investigation in the coordinate system (7.5.8), which is espe-
cially convenient for analyzing density inhomogeneities in an inflationary universe [218,
220]. If, however, one is interested in describing the evolution of the universe from the
point of view of a comoving observer, then it is more convenient to go to a synchronous
coordinate system, which can be so chosen that the metric of an inflationary universe on
scales much larger than H−1 may be written in the form [135, 133]

ds2 ≈ dt2 − a2 (x, t) dx2 , (10.2.1)

where Z t 
a(x, t) ∼ exp H[ϕ(x, t)] dt . (10.2.2)
0

What these expressions mean is that an inflationary universe in a neighborhood of size

l> −1
∼ H about every point x looks like a homogeneous inflationary universe with Hubble
parameter H[ϕ(x, t)]. To study the global structure of the inflationary universe in this
 208

approximation, it turns out to be sufficient to study the independent local evolution

(in accordance with the “no hair” theorem in de Sitter space) of the field ϕ within each
individual region of the inflationary universe of size l ∼ H−1 (or with initial size l0 ∼ H−1 ),
and then attempt to discern the overall picture using Eqs. (10.2.1) and (10.2.2). The local
evolution of the field ϕ in regions whose size is of order H−1 is governed by the diffusion
equations (7.3.22), (7.4.4), and (7.4.5), taking into consideration the dependence of the
H3
diffusion and mobility coefficients D = on the magnitude of the field ϕ [135, 133].
8 π2
The simplest possibility would be to study stationary solutions of Eqs. (7.4.4) and
(7.4.5). The corresponding solution in the general case might also depend on the station-
ary probability flux jc = const, and for V(ϕ) ≪ M4P , it is given by [135]
q
3 M4P 6 π V(ϕ)
" #
Pc (ϕ) ∼ const · exp − 2 jc . (10.2.3)
8 V(ϕ) MP V′ (ϕ)
Unfortunately, attempts to provide a physical interpretation of this solution meet with
all sorts of difficulties. As in Section 7.4, let us first consider the case jc = 0. One
can readily show that Eq. (10.2.3) is identical to the square of the Hartle–Hawking
wave function (10.1.17), namely (10.1.18). But since the effective potential V(ϕ) vanishes
at its minimum, which corresponds to the vacuum state in the observable part of the
universe, the distribution (10.2.3) is unnormalizable. This is an especially easy problem
to comprehend in the chaotic inflation scenario for theories with V(ϕ) ∼ ϕ2 n ; in these
theories, inflation takes place only when ϕ > ∼ MP . In such theories, then, there is no
diffusion flux out of the region with ϕ ∼ MP and into the region with ϕ >
< ∼ MP . But such
a flux would provide the only way to compensate for the classical rolling down of the field
to the minimum of V(ϕ), which is necessary for the existence of a stationary distribution
Pc (ϕ) when jc = 0.
The issue of the interpretation of the second term in (10.2.3) is even more complicated.
As we noted in Section 7.4, formally, in theories with V(ϕ) ∼ ϕ2 n , this solution does not
exist in general, as it is odd in ϕ while Pc (ϕ) must always be positive. We can avert this
problem to a certain extent by recalling that in these theories, Eq. (7.4.5) itself holds
4
only on the segment MP < ∼ ϕ <∼ ϕP , where V(ϕP ) ∼ MP . But this is not an entirely
satisfactory response. It is actually straightforward to show that the second term in
(10.2.3) is a solution of Eq. (7.3.22) in which the first (diffusion) term is omitted. Thus,
we are simply dealing with a classical “rolling” of the field ϕ away from a region with
V(ϕ) ≫ M4P , where the diffusion equation is not valid. In this instance, a stationary
distribution Pc (ϕ, t) can only be sustained through a constant flux jc out of a region with
V(ϕP ) ≫ M4P . One could attempt to interpret this flux as a current corresponding to the
4
probability of quantum creation of new domains of the universe with V(ϕP ) > ∼ MP , per
unit initial coordinate volume. But as Starobinsky has already emphasized in [135], where
a solution of the type (10.2.3) was first derived, at present we can neither give a rigorous
existence proof for such a solution, nor can we say anything definite about the magnitude
of jc if it is nonzero. The feasibility of the interpretation suggested above does not follow
from the derivation of Eqs. (7.4.4) and (7.4.5) given in [132–135]. Moreover, since the
 209

vast preponderance of the initial coordinate volume of the inflationary universe eventually
4
transforms into a state with ϕ < ∼ MP , V(ϕP ) ≪ MP , the validity of the assumed constancy
of the probability current for quantum creation of the new domains of the universe per
unit initial volume seems not to be well-founded. It is not entirely clear, for example,
why it is necessary to require a stationary distribution Pc (ϕ, t) rather than a probability
distribution for finding the field ϕ at time t, per unit physical volume, and taking into
account the increase in volume due to the quasiexponential expansion of the universe
(10.2.2), which proceeds at different rates in regions filled with different fields ϕ.
One can attain a proper understanding of the situation for stationary solutions only
through a comprehensive analysis of nonstationary solutions of the diffusion equation
under the most general initial conditions for the distribution Pc (ϕ, t). As stated above,
we are interested in the probability Pc (ϕ, t) of finding the field ϕ at time t in a region of
initial size O(H−1 ). Due to inflation, the initial inhomogeneities of the field ϕ on this scale
become exponentially small, while the amplitude of quasiclassical perturbations δϕ with
wavelengths l > −1
∼ H on this scale do not exceed H; see (7.3.12). Bearing in mind that
inflation occurs in theories with V(ϕ) ∼ ϕ2 n when ϕ > ∼ MP , the condition V(ϕ) ≪ MP
4

implies that δϕ ∼ H ≪ ϕ, or in other words the field ϕ is, to a very good approximation,
homogeneous on a scale l ∼ H−1 .
We may therefore assume without loss of generality that at time t = 0, the field ϕ
is equal to some constant ϕ0 in the region under consideration — the size of which is
O(H−1 ) — or in other words, Pc (ϕ, t = 0) = ϕ(ϕ − ϕ0 ). Solutions of Eq. (7.3.22) with
such initial conditions were investigated in [132, 133], and there it was found that these
solutions are all nonstationary. The distribution Pc (ϕ, t) first broadens, and its center
is then displaced toward small ϕ, being governed by the same law as the classical field
ϕ(t). Meanwhile, the distribution Pp (ϕ, t) of the physical volume occupied by the field
ϕ behaves differently depending on the value of the initial field ϕ = ϕ0 . For small ϕ0 ,
Pp (ϕ, t) behaves in almost the same way as Pc (ϕ, t), but for sufficiently large ϕ0 , the
distribution Pp (ϕ, t) starts to edge toward large values of the field ϕ as t increases, which
leads to the onset of the self-regenerating inflationary regime discussed in Section 1.8.
Referring the reader to [132, 133] for details, let us elucidate the behavior of the
λ
distributions Pc (ϕ, t) and Pp (ϕ, t) using the theory V(ϕ) = ϕ4 as an example. In order
4
to do so, we divide the quasiclassical field ϕ into a homogeneous classical field ϕ(t) and
inhomogeneities δϕ(x, t) with wavelengths l > −1
∼ H (see (7.5.7)):
ϕ(x, t) = ϕ(t) + δϕ(x, t) . (10.2.4)

It can readily be shown that during the inflationary stage, the equations of motion for
ϕ(t) and δϕ in the metric (10.2.1), (10.2.2) are given, to terms linear in δϕ, by

dV
3 H ϕ̇ = − = −λ ϕ3 , (10.2.5)
dϕ
(V′ )2
" #
1 5
3 H δ ϕ̇ − 2 ∆δϕ = V −′′
δϕ = − λ ϕ2 δϕ . (10.2.6)
a 2V 2
 210

(V′ )2
The term δϕ in (10.2.6) appears by virtue of the dependence of the Hubble param-
2V
eter H on ϕ. Equations (10.2.5) and (10.2.6) make it clear that to lowest order in δϕ, an
investigation of the evolution of the field ϕ(x, t) reduces to an investigation of the motion
of the homogeneous field ϕ(t) as governed by the classical equation of motion (10.2.5),
and to a subsequent investigation of the evolution of the distribution Pc (δϕ, t) subject to
the initial condition Pc (δϕ, 0) ∼ δ(δϕ).
It is important that when ϕ ≫ MP (that is, during inflation), the effective mass
squared of the field δϕ,
(V′ )2 5
m2δϕ = V′′ − = λ ϕ2 , (10.2.7)
2V 2
be much less than the square of the Hubble parameter: m2δϕ ≪ H2 . This means that during
the first stage of “broadening” of the delta functional distribution Pc (δϕ, 0) ∼ δ(δϕ), right
up to the time
3H √
t1 ∼ 2
∼ (2 λ MP )−1 ,
2 mδϕ
the dispersion squared of fluctuations grows linearly with time, (7.3.12):
√
2 H3 (ϕ) t λ λ ϕ6
hδϕ i = = √ t. (10.2.8)
4 π2 3 6 π M3P

The growth of hδϕ2 i then slows down (see (7.3.13)), and by

√
6π
t2 ∼ √ ∼ 10 t1 ,
λ MP
the dispersion of fluctuations δϕ will have essentially reached its asymptotic value (7.3.3):
v s
3 H4 λ ϕ3
q u
u
∆0 = hδϕ2 i = Ct ≈ , (10.2.9)
8 π 2 m2δϕ 15 M2P

where C ≈ 1. Equation (1.7.22) tells us that at this stage (with t ≪ t2 ), the mean field ϕ(t)
hardly decreases at all. For t > t2 , both the field ϕ(t) and the quantity H(ϕ) start to fall
off rapidly, and that is why fluctuations
q produced at t ≫ t2 make a negligible contribution
2
to the total dispersion ∆(t) = hδϕ i. The latter quantity is basically determined by
fluctuations that make their appearance when t < ∼ t2 . To analyze the behavior of ∆(t)
for t > t2 , it is sufficient to note that the amplitude of fluctuations δϕ that arise when
t < t2 subsequently behaves in the same way as the magnitude of ϕ̇ [114] (the reason
dϕ
being that, as one can readily prove, ϕ̇ = obeys the same equation of motion (10.2.6)
dt
as δϕ). This implies that for t ≫ t2 ,

ϕ̇(t) ϕ̇(t)
∆(t) ≈ ∆0 ≈ ∆0 . (10.2.10)
ϕ̇(t2 ) ϕ̇(0)
 211

λ 4
In the theory with V(ϕ) = ϕ , Eq. (1.7.22) implies that ϕ̇ ∼ ϕ(t); that is, for t ≫ t2 ,
4
s
λ ϕ(t) ϕ20
∆(t) = C , (10.2.11)
15 M2P
with C ≈ 1.
The foregoing is an elementary derivation of the expressions (10.2.8)–(10.2.11) for the
dispersion ∆(t), as an attempt to elucidate the physical nature of the phenomena taking
place [57, 78]. The same results can be obtained in a more formal manner by solving the
diffusion equation (7.3.22) directly for the distribution Pc (ϕ, t) with the initial condition
Pc (ϕ, t) ∼ δ(ϕ − ϕ0 ). This problem has been solved in [132, 133]; here we simply present
λ ϕn
the final result for ∆(t) in theories with V(ϕ) = :
n MPn−4

4 λ ϕn−2(t) 4
∆2 (t) = [ϕ0 − ϕ4 (t)] . (10.2.12)
3 n2 MnP
λ 4
In particular, in a theory with V(ϕ) = ϕ,
4
s
1 λ ϕ(t) 4
∆(t) = 2
[ϕ0 − ϕ4 (t)]1/2 . (10.2.13)
2 3 MP
Using (1.7.21), one can easily show that this result is consistent with Eqs. (10.2.8)–
(10.2.11), which were obtained by a simpler method.
In what follows, it will be especially important to analyze the evolution of the scalar
field ϕ during the initial stage of the process (for a time t < ∼ t2 ), during which time the
field ϕ(t) changes by an amount ∆ϕ < ∼ 0ϕ . We already know from (10.2.12) that at that
4
stage ∆(t) ≪ ϕ(t) if V(ϕ0 ) ≪ MP . Thus, if the initial energy density is much less than the
Planck density, the dispersion of the scalar field distribution during the stage in question
will always be much less than the mean value of the field ϕ(t) — that is, we are justified
in investigating the evolution of the field ϕ(x, t) to first order in δϕ(x, t) (10.2.5), (10.2.6).
On the other hand, when ∆(t) ≪ ϕ(t), Pc (ϕ, t) is a Gaussian distribution in the vicinity
of its maximum at ϕ = ϕ(t), i.e.,

[ϕ − ϕ(t)]2
!
Pc (ϕ, t) ∼ exp −
2 ∆2
3 n2 [ϕ − ϕ(t)]2 MnP
!
= exp − , (10.2.14)
8 λ ϕn−2(t) [ϕ40 − ϕ4 (t)]

where ϕ(t) is a solution of Eq. (10.2.5); see (1.7.21), (1.7.22). In particular,

 s 
λ
ϕ(t) = ϕ0 exp − MP t (10.2.15)
6π
 212

λ 4
in the theory with V(ϕ) = ϕ , and
4
s
n n λ 3− n2
 
2− n 2− n
ϕ 2 (t) = ϕ0 2 −t 2− M (10.2.16)
2 24 π P
λ ϕn
in the theory with V(ϕ) = for n 6= 4.
n MPn−4
In any given domain of the universe, then, the distribution Pc (ϕ, t) is nonstationary,
and in the course of time, the probability of detecting a large field ϕ at any given point
of space becomes exponentially low.
If, however, one would like to know the distribution Pp (ϕ, t) of the physical volume of
the universe, taking account of expansion proportional to
Z t 
exp H(x, t) dt ,
0

that contains the field ϕ at time t, the answer will be completely different. To attack this
question, let us consider for definiteness the evolution of the distribution Pc (ϕ, t) for a
ϕ0
time ∆t, during which the mean field ϕ(t) decreases by ∆ϕ = ≪ ϕ0 , where N is some
N
number satisfying N ≫ 1. According to (10.2.14),
2
1
   
2
 3 n N ϕ − ϕ0 1 − MnP 
Pc (ϕ, ∆t) ≈ exp −
 N 
. (10.2.17)
32 λ ϕn+2

0
 

We see from this result that the fraction of the original coordinate volume remaining in
a state with ϕ = ϕ0 after a time ∆t is

3 n2 MnP 3 n M4P
! !
Pc (ϕ0 , ∆t) ≈ exp − = exp − . (10.2.18)
32 λ N ϕn0 32 N V(ϕ0 )

Notice that Pc (ϕ0 , ∆t) ≪ 1 when V(ϕ0 ) ≪ M4P . In other words, the dispersion ∆ is much
less than the difference ϕ0 − ϕ(t). In a time ∆t, the volume of a region with ϕ = ϕ0
expands by a factor e3 H(ϕ0 ) δt , on the average. It then follows from (10.2.15) and (10.2.16)
that n
s −3
2 6 π MP2
∆t = n . (10.2.19)
N n λ ϕ 2 −2
The original volume occupied by the field ϕ0 thus changes in a time ∆t given by (10.2.19)
by a factor Pp (ϕ0 , ∆t), where

Pp (ϕ0 , ∆t) ≈ Pc (ϕ0 , ∆t) exp[3 H(ϕ0 ) ∆t]

3 n2 MnP 24 π ϕ20
!
= exp − + . (10.2.20)
32 λ N ϕn0 N n M2P
 213

Clearly, then, when ϕ0 ≫ α ϕ∗ , where

n3 ( n+2 )
! 1
1
− n+2
ϕ∗ = λ MP , α= = O(1) , (10.2.21)
28 π

the volume occupied by the field ϕ0 will grow during the time ∆t rather than decrease. The
same thing will be repeated during the next time interval ∆t, and so on. This means that
during inflation, regions of the inflationary universe with ϕ > ϕ∗ reproduce themselves
endlessly: the process of inflation, once having started, continues forever unabated, and
the volume of the inflationary part of the universe grows without bound.
ϕ0
Behavior of the distribution Pp (ϕ, ∆t) for ϕ−ϕ0 ≫ ∆ϕ = is even more interesting.
N
ϕ0
In point of fact, a field ϕ that occasionally jumps to values with ϕ − ϕ0 ≫ due to
N
quantum fluctuations in some domain of the inflationary universe cannot be significantly
reduced in a time ∼ ∆t, either by classical rolling (by ∆ϕ ≪ ϕ − ϕ0 ) or by diffusion (by
∼ ∆ ≪ ∆ϕ). The volume of all regions occupied by the field ϕ increases in a time ∆t
(10.2.19) by a factor exp[3 H(ϕ) ∆t], whereupon

Pp (ϕ, ∆t) ≈ Pc (ϕ, ∆t) exp[3 H(ϕ) ∆t]

 !n/2 
 3 n2 N [ϕ − ϕ(t)]2 MnP 24 π ϕ20 ϕ 
= exp − + .
 32 λ ϕn+2
0 N n M2P ϕ0 

(10.2.22)
1
n2 N n+2
!
It can readily be shown that when ϕ > β ϕ∗ , where β = = O(1), the maximum
26
of the distribution Pp (ϕ, ∆t) is shifted not towards ϕ < ϕ0 , like the maximum of Pc (ϕ, ∆t),
but towards ϕ > ϕ0 .
This means that when ϕ ≫ ϕ∗ , the universe not only continually regenerates itself,
but in the process, most of the physical volume of the universe gradually fills with a larger
and larger field ϕ [132, 133]. This result is entirely consistent with the result obtained in
Section 1.8 by more elementary methods [57].

10.3 The self-regenerating inflationary universe and quantum cosmology

The possibility of an eternally existing, self-regenerating universe is one of the most impor-
tant and surprising consequences of the theory of the inflationary universe, and it merits
detailed discussion (see also Section 1.8). Let us first of all provide a more accurate
interpretation of the results obtained in Section 10.2.
The physical meaning of the distributions Pc (ϕ, t) and Pp (ϕ, t) is as follows. Consider
a domain of the inflationary universe having initial size l > −1
∼ H , and let us assume that
initially (at t = 0) it is uniformly filled throughout its entire volume by observers with
 214

identical clocks, which are synchronized at t = 0. In that event, the quantity Pc (ϕ, t)
determines the fraction of all observers who at time t as measured by their own clocks
(i.e., in a synchronous coordinate system) are located in a region filled with a practically
homogeneous (on a scale l > −1
∼ H (ϕ)) semiclassical field ϕ. The distribution Pp (ϕ, t)
determines how much of the physical volume of the universe is occupied by observers
who, at time t as measured by their own clocks, live in regions filled with the field ϕ that
is homogeneous on a scale larger than H−1 (ϕ).
The results obtained in the preceding section imply that in no particular region of the
inflationary universe can the distribution Pc (ϕ, t) be stationary. It can be quasistationary
during tunneling from a metastable vacuum state to a stable state, as in the case of
Hawking–Moss tunneling in the new inflationary universe scenario. But in any model in
which the universe becomes hot after inflation and the field ϕ rolls down to its minimum
at V(ϕ) = 0, the function Pc (ϕ, t) cannot (and should not) be a nontrivial stationary
distribution (at least within the range of validity of the approximation that we have
employed; see below). In other words, the fraction of all observers initially situated
in an unstable state away from the absolute minimum of the effective potential V(ϕ)
should decrease with time. This conclusion is confirmed by results obtained above — for
example, see Eq. (10.2.14), which shows that the probability of remaining√ in an unstable
λ 4 6π ϕ0
state ϕ > ∼ MP in a theory with V(ϕ) = 4 ϕ after a time t > ∼ √λ M ln M becomes
P
exponentially small.
At the same time, when ϕ0 ≫ ϕ∗ (see (10.2.21)), the distribution Pp (ϕ, t) begins to
increase with increasing ϕ in a region with ϕ > ∗
∼ ϕ , i.e., the fraction of the volume of the
inflationary universe occupied by observers who at time t by their clocks find themselves
located in an unstable state ϕ > ∗
∼ ϕ increases at large ϕ and t, and consequently the total
volume of the inflationary regions of the universe continues to grow without bound. It
follows from (10.2.22) that at large t, most of the volume of the universe should be in a
state with the very largest possible value of the field ϕ, such that V(ϕ) ∼ M4P .
Here, to be sure, we must state an important reservation. The fraction of the volume
of the universe in a state with a given field ϕ at a given time depends on what we mean
by the word time. The results obtained above refer to the proper time t of comoving
observers whose clocks were synchronized at some time t = 0, when they were all quite
close to one another. The same phenomena can be described using another coordinate
system, namely the coordinates (7.5.8), which are especially convenient in investigating
the density inhomogeneities produced during inflation. In order to distinguish between
the proper time t in the synchronous coordinates and the “time” in the coordinates (7.5.8),
we denote the latter here as τ . Investigation of diffusion in the coordinate system (7.5.8)
also shows that the total volume of the universe filled with the field ϕ > ϕ∗ increases
exponentially with time τ [133]. But due to the specific way in which the “time” τ
is defined, the rate of exponential expansion of the universe, which is ∼ eH τ in the
coordinates (7.5.8), is the same everywhere, regardless of any local increases or decreases
in the field ϕ. The fraction of the physical volume of the universe filled with a large field ϕ
on a hypersurface of constant τ therefore falls off in almost the same way as Pc (ϕ, t), and
 215

thus the fraction of the physical volume of a self-regenerating universe that is transformed
over time into a state with the largest possible value of the field ϕ depends on what exactly
one means by the word time. It is for just this reason that we have engaged here in a
more detailed discussion of this question, the answer to which turns out to depend on
exactly how the question is formulated. Fortunately, however, our basic conclusion about
self-regeneration and exponential expansion of regions of the universe filled with a field
ϕ > ϕ∗ is coordinate-independent [133].
It is worthwhile to examine these results from yet another standpoint. If the universe is
self-regenerating, then the standard question about the initial conditions over the whole
universe may be irrelevant, since the universe may turn out not to have had a global
initial spacelike singular hypersurface to play the role of a global Cauchy hypersurface.
At present, we do not have sufficient reason to believe that the universe as a whole was
created approximately 1010 years ago in a singular state, prior to which classical space-time
did not exist at all. Inflation could begin and end at different times in different domains of
the universe, and this would be completely consistent with the existing observational data.
Accordingly, the matter density in different regions of the universe will drop to ρ0 ∼
10−29 g/cm3 at different times, approximately 1010 years after inflation ends in each of
these regions. It is just after this point in each of these regions that the conditions
required for the emergence of observers like ourselves will first appear. The number of
such observers should plainly be proportional to the volume of the universe at the density
hypersurface(s) with ρ = ρ0 ∼ 10−29 g/cm3 . Therefore, having investigated the question
of what processes take part in the creation of most of the volume of the universe at the
density hypersurface ρ = ρ0 ∼ 10−29 g/cm3 (i.e., 1010 years after inflation ceases in each
particular domain), we have prepared ourselves to assess the most likely history of the
part of the universe that we are able to observe.
In order to look into this question, one should take into account that the universe
expands by a factor of approximately 1030 over the 1010 years after the end of inflation,
λ
and that during inflation in the theory with V(ϕ) = ϕ4 , the universe typically expands
4
π ϕ20
!
by a factor exp , where ϕ0 is the original value of the field ϕ. However, for
M2P
ϕ0 >∼ϕ ∼λ
∗ −1/6
MP , this result is modified.
Indeed, as at the end of Section 10.2, let us consider a domain of the inflationary
universe in which the scalar field, due to its long-wave quantum fluctuations during the
time ∆t, jumps from ϕ = ϕ0 up to some value ϕ such that ϕ − ϕ0 ≫ |∆ϕ|, where ∆ϕ
is the value of the classical decrease in the field ϕ during the time ∆t. If the jump in
the field ϕ is large enough, its mean value in this domain will return to the original value
ϕ = ϕ0 , mainly due to classical rolling. According to (1.7.25), during " classical rolling,
# the
π
domain under consideration inflates by an additional factor of exp (ϕ2 − ϕ20 ) .
M2P
The probability of a large jump in the field ϕ is exponentially suppressed (see (10.2.14),
(10.2.22)), but it is not hard to prove that when ϕ ≫ ϕ∗ , this suppression pays us back
with interest on account of the aforementioned additional inflation of the region filled with
 216

the field ϕ making the jump. This means that most of the volume of the universe after
inflation (for example, on the hypersurface ρ = ρ0 ) results from the evolution of those
relatively rare but additionally inflated regions in which the field ϕ has jumped upward
as a result of long-wavelength quantum fluctuations. To continue this line of reasoning, it
can be shown that the overwhelming preponderance of the physical volume of the universe
in a state with given density ρ = ρ0 is formed as a result of the inflation of regions in
which the field ϕ, over the longest possible times, has been fluctuating about its maximum
possible values, such that V(ϕ) ∼ M4P . In that sense, a state with potential energy density
close to the Planck value (i.e., the space-time foam) can be considered to be a source,
continuously producing the greater part of the physical volume of the universe. We shall
return to this point subsequently, but for the moment we wish to compare our conclusions
with the basic expectations and assumptions that have been made in analyzes of the wave
function of the universe.
In deriving the expression for the wave function (10.1.12), (10.1.17) proposed by Har-
tle and Hawking, it was assumed that the universe has a stationary ground state, or a
state of least excitation (vacuum), the wave function Ψ(a, ϕ) of which they attempted to
determine; see Section 10.1. The square of this wave function |Ψ(a, ϕ)|2 should then give
the stationary distribution for the probability of detecting the universe in a state with
the homogeneous scalar field ϕ and scale factor a (10.1.18), (10.1.20). The fact that the
quasistationary distribution Pc (ϕ) (10.2.3) is proportional to the square of the Hartle–
Hawking wave function could be taken as a important indication of the validity of this
assumption. The results obtained in the preceding section, however, show that with fairly
general initial conditions, the distribution Pc (ϕ, t) in the inflationary universe scenario
does not wind up in the stationary regime (10.2.3). Nevertheless, another type of station-
ary regime is possible, described in part by the distribution Pp (ϕ, t). In this regime, the
universe continually produces exponentially expanding regions (mini-universes) contain-
4
ing the large field ϕ (with ϕ∗ <∼ϕ< ∼ ϕP ), where V(ϕP ) ∼ MP , and the properties of the
universe within such regions do not depend on the properties of neighboring regions (by
virtue of the “no hair” theorem for de Sitter space), nor do they depend on the history or
epoch of formation of those regions. Here we can speak of stationarity in the sense that
regions of the inflationary universe containing a field ϕ > ∗
∼ ϕ constantly come into being,
and in an exponentially large neighborhood of each such region, the average properties
of the universe are the same and do not depend on the epoch at which the region was
formed. This implies that the inflationary universe has a fractal structure [133, 325].
Thus, the Hartle–Hawking wave function may be useful in describing the intermediate
stages of inflation, during which the distribution Pc (ϕ, t) can sometimes (in the presence
of metastable vacuum states) turn out to be quasistationary. This function may also
turn out to be quite useful in the study of certain other important problems of quantum
cosmology — see, for example, a discussion of this possibility in Section 10.7. At the
present time, however, we are unable to ascribe to this wave function the fundamental
significance sometimes assigned to it in the literature.
What then can we say about the tunneling wave function (10.1.23) used to describe
the quantum creation of the universe?
 217

To answer this question, let us study in somewhat more detail the first (diffusive)
stage of spreading of the initial distribution Pc (ϕ, 0) = δ(ϕ − ϕ0 ) when the magnitude
of the classical field ϕ is almost constant, ϕ(t) ≈ ϕ0 . Equation (10.2.14) holds when the
dispersion ∆ and the difference between ϕ and ϕ0 are much less than ϕ0 itself. At the
same time, when ϕ − ϕ0 ∼ ϕ0 the distribution Pc (ϕ, t) is far from Gaussian. In order to
calculate in that case, one should bear in mind that the classical rolling of the field ϕ, i.e.,
the last term in the diffusion equation (7.3.22), can be neglected during the first stage:
√
∂Pc (ϕ, t) 2 2 ∂2
= √ [V(ϕ) Pc (ϕ, t)] . (10.3.1)
∂t 3 3 π M3P ∂ϕ2
It is convenient to seek a solution of this equation in the form
" #
S(ϕ)
Pc (ϕ, t) ∼ A(ϕ, t) · exp − ,
t

where A(ϕ, t) and S(ϕ) are relatively slowly varying functions of ϕ and t.
λ ϕn
It can readily be shown that in the theory V(ϕ) = with ϕ ≪ ϕ0 , the corre-
n MPn−4
sponding solution is
 √ ! 3n −1 
3 6π MP 2
Pc (ϕ, t) = A exp − √  , (10.3.2)
t λ λ (3 n − 4)2 ϕ

and neglect of the last term in (7.3.22) is warranted when

s ! n −2
6 π −1 MP 2
t<
∼ ∆t(ϕ) = M (10.3.3)
nλ P ϕ

(compare with (10.2.19)). If the effective potential V(ϕ) is not too steep (n ≤ 4), then the
diffusion approximation first ceases to work at small ϕ, and subsequently at ϕ ∼ ϕ0 . One
can say in that event that regions of space with a small field ϕ, in which classical motion
prevails over quantum fluctuations, are formed by virtue of a quantum diffusion process
that operates for a time t <∼ ∆t(ϕ), and the probability distribution for the creation of
a region (mini-universe) with a given field ϕ when quantum diffusion ceases to dominate
(t ∼ ∆t(ϕ)) is, according to (10.3.2) and (10.3.3),

M4P
" #
Pc (ϕ, ∆t(ϕ)) ∼ exp −C , (10.3.4)
V(ϕ)

where C = O(1). This formula holds for n ≤ 4, ϕ ≪ ϕ0 regardless of the initial value of
4
the field ϕ0 . In particular, when V(ϕ0 ) >
∼ MP , it can be interpreted as the probability for
4
the quantum creation of a (mini-)universe from the space-time foam with V(ϕ) > ∼ MP .
It is easily seen that up to a factor C = O(1), (10.3.4) and the probability (10.1.24) of
quantum creation of a universe from “nothing” are identical.
 218

Is this merely formal consistency between these equations, or is there more to it than
that? An answer requires additional investigation, but there are some ideas on this score
that may be enunciated right now. First of all, note that rather than describing the
creation of the entire universe from “nothing,” Eq. (10.3.4) describes the creation of
only a part of the universe of size greater than H−1 (ϕ) due to quantum diffusion from a
previously existing region of the inflationary universe. Moreover, Eq. (10.3.4) holds in
theories with V(ϕ) ∼ ϕn only when n ≤ 4; for n > 4, it can be shown that
 ! n −2 
C M4P ϕ0 2
Pc (ϕ, ∆t(ϕ)) = Pc (ϕ, ∆t(ϕ0 )) ∼ exp −  . (10.3.5)
V(ϕ) ϕ

In theories in which the interval between ϕ and ϕ0 contains segments where the field ϕ
rolls down rapidly and the universe is not undergoing inflation, equations like (10.3.4)
and (10.3.5) will generally not be valid; that is, the diffusion equation that we have used
will not be applicable to such segments. More specifically, one cannot derive equations
like (10.3.4) for the probability of diffusion from a space-time foam with V(ϕ0 ) ∼ M4P
onto the top of the effective potential at ϕ = 0 in the new inflationary universe scenario.
Meanwhile, it is usually assumed that Eq. (10.1.23) (perhaps somewhat modified to take
account of the effects of quantum creation of particles during tunneling [321]) can describe
the quantum creation of the universe as a whole, even if a continuous diffusive transition
between ϕ0 and ϕ is not possible.
This indicates that we are dealing here with two distinct complementary or compet-
ing processes described by Eqs. (10.3.4) and (10.1.24), respectively. However, experience
with the Hawking–Moss theory of tunneling (7.4.1) engenders a certain amount of caution
in this regard. Recall that Eq. (7.4.1), which was originally derived via the Euclidean
approach to tunneling theory, was interpreted as giving the probability of uniform tun-
neling over the entire universe [121]. However, a rigorous derivation of Eq. (7.4.1) and
a justification for this interpretation were lacking. In our opinion, a rigorous derivation
of Eq. (7.4.1) was first provided by solving the diffusion equation (7.3.22), and its inter-
pretation was different from the original one based on the Euclidean approach [134, 135],
though consistent with the interpretation proposed in [209]. Likewise, neither approach
to deriving Eq. (10.1.24) (using the (anti-)Wick rotation t → i τ or considering tunneling
from the point a = 0) is sufficiently rigorous, and the interpretation of (10.1.24) as the
probability of tunneling from “nothing” also falls somewhere on the borderline between
physics and poetry. One of the fundamental questions to emerge here had to do with what
exactly was tunneling, if there were no incoming wave. A plausible response is that one
simply cannot identify the incoming wave within the framework of the minisuperspace
approach. Indeed, by solving the diffusion equation in the theory of chaotic inflation,
it was shown that during inflation there is a steady process of creation of inflationary
domains whose density is close to the Planck density, and whose size is l ∼ lP ∼ M−1 P .
Tunneling (or diffusion) involving an increase in the size of each such region and a change
in the magnitude of the scalar field within each region can be (approximately) associated
with the process of quantum creation of the universe. The process of formation of such
 219

Planck-size inflationary domains (the “incoming wave”) cannot be described in the con-
text of the minisuperspace approach, but it has a simple interpretation within the scope
of the stochastic approach to inflation.
We have thus come closer to substantiating the validity of Eq.
(10.1.24) as a probability of quantum creation of the universe from “nothing.” It is
nevertheless still not entirely clear whether in this expression there is anything that is
both true and at the same time different from Eq. (10.3.4), which was derived with the
aid of the stochastic approach to inflation, and which has a much more definite physical
meaning. This is a particularly important question, as it pertains to the theory of the
quantum creation of the universe in the state ϕ = 0, corresponding to a local maximum
of V(ϕ) located at V(ϕ) ≪ M4P , since diffusion into this state from a space-time foam
with V(ϕ) ∼ M4P is impossible.
To conclude this section, let us examine one more question, relating to the possibility
of creating an inflationary universe from Minkowski space. The issue here is that quantum
fluctuations in the latter can bring into being an inflationary domain of size l > −1
∼ H (ϕ),
where ϕ is a scalar field produced by quantum fluctuations in this domain. The “no hair”
theorem for de Sitter space implies that such a domain inflates in an entirely self-contained
manner, independent of what occurs in the surrounding space. We could then conceive
of a ceaseless process of creation of inflationary mini-universes that could take place even
at the very latest stages of development of the part of the universe that surrounds us.
A description of the process whereby a region of the inflationary universe is produced
as a result of quantum fluctuations could proceed in a manner similar to that for the
formation of regions of the inflationary universe with a large field ϕ through the buildup
of long-wave quantum fluctuations δϕ. The basic difference here is that long-wave fluctu-
ations δϕ of the massive scalar field ϕ at the time of inflation with m ≪ H are “frozen”
in amplitude, while there is no such effect in Minkowski space. But if the buildup of
quantum fluctuations in some region of Minkowski space were to engender the creation
of a fairly large and homogeneous field ϕ, then that region in and of itself could start to
inflate, and such a process could stabilize (“freeze in”) the fluctuations δϕ that led to its
onset. In that event, one could sensibly speak of a self-consistent process of formation of
inflationary domains of the universe due to quantum fluctuations in Minkowski space.
Without pretending to provide a complete description of such a process, let us attempt
λ ϕn
to estimate its probability in theories with V(ϕ) = . A domain formed with a large
n MPn−4
field ϕ will only be a part of de Sitter space if in its interior (∂µ ϕ)2 ≪ V(ϕ). This means
that the size of the domain must exceed l ∼ ϕ V(ϕ)−1/2 , and the field inside must be
greater than MP . Such a domain could arise through the buildup of quantum fluctuations
δϕ with a wavelength
k −1 >
∼ l ∼ ϕV
−1/2
(ϕ) ∼ m−1 (ϕ) .
One can estimate the dispersion hϕ2 ik<m of such fluctuations using the simple formula
1 Z m(ϕ) k 2 dk m2 V(ϕ)
hϕ2 ik<m ∼ 2
q ∼ 2
∼ 2 2 , (10.3.6)
2π 0 k 2 + m2 (ϕ) π π ϕ
 220

and for a Gaussian distribution P(ϕ) for the appearance of a field ϕ which is sufficiently
homogeneous on a scale l, one has [133]

π 2 ϕ4
" #
P(ϕ) ∼ exp −C , (10.3.7)
V(ϕ)

λ 4
where C = O(1). In particular, for a theory with V(ϕ) = ϕ ,
4
4 π2
" #
P(ϕ) ∼ exp −C . (10.3.8)
λ

Naturally, this method is rather crude; nevertheless, the estimates that it provides are
quite reasonable, to order of magnitude. For example, practically the same lines of rea-
soning could be employed in assessing the probability of tunneling from the point ϕ = 0
λ
in a theory with V(ϕ) = − ϕ4 ; see Chapter 5. The estimate that one obtains for the
4
formation of a bubble of the field ϕ is also given by Eq. (10.3.8). This result is in complete
8 π2
!
accord with the equation P ∼ exp − (5.3.12), which was derived using Euclidean
3λ
methods. In fact, one can easily verify that all results concerning tunneling which were
obtained in Chapter 5 can be reproduced (up to a numerical factor C = O(1) in the
exponent) by using the simple method suggested above. This makes the validity of the
estimates (10.3.7), (10.3.8) quite plausible.
The main objection to the possibility of quantum creation of an inflationary universe
in Minkowski space is that energy conservation forbids the production of an object with
positive energy out of the vacuum in this space. Within the scope of classical field theory,
in which the energy is everywhere positive, such a process would therefore be impossible
(a related problem is discussed in [213, 326]). But at the quantum level, the energy
density of the vacuum is zero by virtue of the cancellation between the positive energy
density of classical scalar fields, along with their quantum fluctuations, and the negative
energy associated with quantum fluctuations of fermions, or the bare negative energy of
the vacuum. The creation of a positive energy-density domain through the buildup of
long-wave fluctuations of the field ϕ is inevitably accompanied by the creation of a region
surrounding that domain in which the long-wave fluctuations of the field ϕ are suppressed,
and the vacuum energy density is consequently negative. Here we are dealing with the
familiar quantum fluctuations of the vacuum energy density about its zero point. It is
important here that from the point of view of an external observer, the total energy of the
inflationary region of the universe (and indeed the total energy of the closed inflationary
universe) does not grow exponentially; the region that emerges forms a universe distinct
from ours, to which it is joined only by a connecting throat (wormhole) which, like a
black hole, can disappear by virtue of the Hawking effect [327, 213]. At the same time,
the shortfall of long-wave fluctuations of the field ϕ surrounding that region is quickly
replenished by fluctuations arriving from neighboring regions, so the negative energy of
 221

the region near the throat can be be rapidly spread over a large volume around the
inflationary domain.
Our discussion of the creation of an inflationary universe in
Minkowski space is highly speculative, and is only intended to illustrate the basic fea-
sibility of such a process; this is clearly a problem that requires closer study. If this
process can actually transpire, and if it is accompanied by burnout of the wormhole con-
necting the parent (Minkowski) space with the inflationary universe that is its offspring,
then the theory will have one more mode for the stationary production of regions of an
inflationary universe. We wish to emphasize, however, that in our approach, the likeli-
hood of this regime being realized is in no way related to the distribution Pc (ϕ), which is
proportional to the square of the wave function (10.1.17). The Euclidean approach to the
theory of baby-universe formation has been developed in [350–352], and will be discussed
in Section 10.7.

10.4 The global structure of the inflationary universe and the problem of the
general cosmological singularity
One of the most important consequences of the inflationary universe scenario is that
under certain conditions, once a universe has come into being it can never again collapse
as a whole and disappear completely. Even if it initially resembles a homogeneous closed
Friedmann universe, it will most likely cease to be locally homogeneous and become
markedly inhomogeneous on the largest scales, and there will then be no global end of
the world such as that which occurs in a homogeneous closed Friedmann universe.
There are versions of the theory of the inflationary universe in which self-regeneration
of the universe does not take place, such as the Shafi–Wetterich model [237], which is
based on a study of inflation in a particular version of Kaluza–Klein theory — see Section
9.5. But for most of the inflationary models studied thus far, the evolution of the universe
and the process of inflation has no end. In the old Guth scenario, for example, when the
probability of forming bubbles of a new phase with ϕ 6= 0 is low enough, these bubbles
will never fill the entire physical volume of the universe, since the distance separating
any two of them increases exponentially, and the resulting increase in the volume of the
universe in the state ϕ = 0 is greater than the decrease of this volume due to the creation
of new bubbles [53, 113, 327, 328]. One encounters a similar effect in the new inflationary
universe scenario [266, 267], and a detailed theory of this process [204] is quite similar to
the corresponding theory in the chaotic inflation scenario [57, 133], the basic difference
being that in both the old and new inflationary universe scenarios one is dealing with the
production of regions containing a field ϕ close to zero and with V(ϕ) ≪ M4P , while in the
chaotic inflation scenario there can be a steady output of regions with very high values
of V(ϕ), right up to V(ϕ) ∼ M4P . We shall subsequently find this to be a very important
circumstance.
The possibility of the ceaseless regeneration of inflationary regions of the universe,
which implies the absence of a general cosmological singularity (i.e., a global spacelike
 222

singular hypersurface) in the future, compels us to reconsider the problem of the initial
cosmological singularity as well. At present, it seems unnecessary to assume that there
was a unique beginning to this endless production of inflationary regions. Models of a
nonsingular universe based on this idea have been proposed in the context of both the old
[327, 328] and new [267] inflationary universe scenarios. According to these models, most
of the physical volume of the universe remains forever in a state of exponential expansion
with ϕ ≈ 0, engendering ever newer exemplars of our type of mini-universe.
Unfortunately, it is not yet entirely clear how one would go about implementing this
possibility. To understand where the main difficulty lies, recall that exponentially ex-
panding (flat) de Sitter space is not geodesically complete — it comprises only a part of
the closed de Sitter space, which originally contracts (at t < 0) rather than expanding:
a(t) = H−1 cosh H t
(see Section 7.2). In de Sitter space contracting at an exponential rate, a phase transition
from the state ϕ = 0 can in principle take place in a finite time over the entire volume,
and there would then remain no regions that could lead to an infinite expansion of the
universe for t > 0. This question, and the problem of the geodesic completeness of a
self-regenerating inflationary universe, requires further study, both because the theory
of phase transitions in an exponentially contracting space is not well understood, and
because the global geometry of a self-regenerating universe differs from the geometry of
de Sitter space. At present, therefore, we cannot say with absolute certainty that the new
inflationary universe scenario with no initial singularities is impossible. That there be no
singularities in the past, however, is probably too strong a requirement.
A more natural possibility is suggested by the chaotic inflation scenario, where most
of the physical volume of the universe at a hypersurface of given density is comprised
of regions which, by virtue of fluctuations in the field ϕ, passed through a stage with
V(ϕ) ∼ M4P . In this scenario, classical space-time behaves as if it were in a state of
dynamic equilibrium with the space-time foam: regions of classical space are continually
created out of the space-time foam, and some of these are reconverted to the foam with
4
V(ϕ) > ∼ MP . In that sense, the occurrence of spatial “singularities” is part and parcel
of this scenario. At the same time, what is graphically clear about this scenario is that
instead of dealing with the problem of creation of an entire universe from a singularity,
prior to which nothing existed, and its subsequent dénouement into nothingness, we are
simply concerned with an endless process of interconversion of phases in which quantum
fluctuations of the metric are either large or small. Our results imply that once it has
arisen, classical space-time — a phase in which quantum fluctuations of the metric are
small — will never again disappear. Even more than in the new inflationary universe
scenario, the global geometric properties of a region filled with this phase3 differ from
those of de Sitter space. If this region turns out to be geodesically complete, then one
can plausibly discuss a model in which the universe has neither a unique beginning nor a
unique end.
3
We traditionally refer to this particular region as the universe, although strictly speaking, the universe
(i.e., all that exists) includes those regions occupied by the space-time foam as well.
 223

Actually, however, as we have already pointed out, this possibility arises in the chaotic
inflation scenario without even taking the self-regeneration process into consideration.
−1
Specifically, if the universe is finite and initially no larger than the Planck size, l <
∼ MP ,
then it is not unreasonable to suppose that at some initial time t = 0 (to within perhaps
∆t ∼ M−1 P ) the entire universe came into being as a whole out of the space-time foam (in
classical language, it appeared from a singularity). If, however, the universe is infinite,
then the possibility that an infinity of causally disconnected regions of classical space will
simultaneously appear out of the space-time foam seems totally unlikely.4
To avoid confusion, it should again be emphasized that the existence of an initial
general (global) spacelike cosmological singularity is not, in and of itself, a necessary
consequence of the general topological theorems on singularities. This conclusion is based
primarily on the assumption of global homogeneity of the universe. Within the framework
of the hot universe theory, such an assumption, even though it had no fundamental
justification to back it up, nevertheless seemed unavoidable, since in that theory the
observable part of the universe arose by virtue of the expansion of an enormous number
of causally disconnected regions in which for some unknown reason the matter density
was virtually the same (see the discussion of the homogeneity and horizon problems in
Chapter 1). On the other hand, in the inflationary universe scenario, the assumption that
the universe is globally homogeneous is unnecessary, and in many cases it is simply wrong.
Therefore, in the context of inflationary cosmology, the conventional statement that in
the very early stages of evolution of the universe there was some instant of time before
which there was no time at all (see Section 1.5) is, at the very least, not well-founded.

10.5 Inflation and the Anthropic Principle

One of the main desires of physicists is to construct a theory that unambiguously predicts
the observed values for all parameters of all the elementary particles that populate our
universe. The noble idealism of the researcher compels many to believe that the correct
theory describing our world should be both beautiful and unique. This does not at all
imply, however, that all parameters of elementary particles in such a theory must be
uniquely calculable. For example, in supersymmetric SU(5) theory, the effective potential
V(Φ, H) for the Higgs fields Φ and H that figure in this theory has several different minima,
and without taking gravitational effects into account, the vacuum energy V(Φ, H) would
be the same at all of these minima. Each of the minima corresponds to a different type
of symmetry breaking in this theory, i.e., to different properties of elementary particles.
Gravitational interactions remove the energy degeneracy between these minima. But
the lifetime of the universe in a state corresponding to any such minimum turns out
either to be infinite or at least many orders of magnitude greater than 1010 years [329].
4
From this standpoint, open and flat Friedmann models, which are quite useful in the description of
local properties of our universe at any stage of its existence. At the same time, the model of a closed
Friedmann universe can describe the global properties of the universe, but only during the earliest stages
of its evolution, until diffusion of the field ϕ results in a large distortion of the original metric.
 224

This means that prescribing a specific grand unified theory will not always enable one
to uniquely determine properties of elementary particles in our universe. An even richer
spectrum of possibilities comes to the fore in Kaluza–Klein and superstring theories, where
an exponentially large variety of compactification schemes is available for the original
multidimensional space; the type of compactification determines the coupling constants,
the vacuum energy, the symmetry breaking properties in low-energy elementary particle
physics, and finally, the effective dimensionality of the space we live in (see Chapter 1).
Under these circumstances, the most varied sets of elementary-particle parameters (mass,
charge, etc.) can appear. It is conceivable that this is the very reason why we have not
yet been able to identify any particular regularity in comparisons of the electron, muon,
proton, W-meson, and Planck masses. Most of the parameters of elementary particles look
more like a collection of random numbers than a unique manifestation of some hidden
harmony of Nature. Meanwhile, it was pointed out long ago that a minor change (by
a factor of two or three) in the mass of the electron, the fine-structure constant αe , the
strong-interaction constant αs , or the gravitational constant would lead to a universe in
which life as we know it could never have arisen. For example, increasing the mass of
the electron by a factor of two and one-half would make it impossible for atoms to exist;
multiplying αe by one and one-half would cause protons and nuclei to be unstable; and
more than a ten percent increase in αs would lead to a universe devoid of hydrogen.
Adding or subtracting even a single spatial dimension would make planetary systems
impossible, since in space-time with dimensionality d > 4, gravitational forces between
distant bodies fall off faster than r −2 [330], and in space-time with d < 4, the general
theory of relativity tells us that such forces are absent altogether.
Furthermore, in order for life as we know it to exist, it is necessary that the universe
be sufficiently large, flat, homogeneous, and isotropic. These facts, as well as a number of
other observations and remarks, lie at the foundation of the so-called Anthropic Principle
in cosmology [77]. According to this principle, we observe the universe to be as it is because
only in such a universe could observers like ourselves exist. There are presently several
versions of this principle extant (the Weak Anthropic Principle, the Strong Anthropic
Principle, the Final Anthropic Principle, etc.) — see [331]. All versions, formulated in
markedly different ways, in one way or another interrelate the properties of the universe,
the properties of elementary particles, and the very fact that mankind exists in this
universe.
At first glance, this formulation of the problem looks to be faulty, inasmuch as
mankind, having appeared 1010 years after the basic features of our universe were laid
down, could in no way influence either the structure of the universe or the properties of
the elementary particles within it. In reality, however, the issue is not one of cause and
effect, but just of correlation between the properties of the observer and the properties
of the universe that he observes (in the same sense as in the Einstein–Podolsky–Rosen
experiment [332], where there is a correlation between the states of two different particles
but no interaction between them). In other words, what is at issue is the conditional
probability that the universe will have the properties that we observe, with the obvious
and apparently trivial condition that observers like ourselves, who take an interest in the
 225

structure of the universe, do indeed exist.

All this discussion can make sense only if one can actually compare the probabilities
of winding up in different universes having different properties of space and matter, but
that is possible only if such universes do in fact exist. If it is not so, any talk of altering
the mass of the electron, the fine structure constant, and so forth is perfectly meaningless.
One possible response to this objection is that the wave function of the universe
describes both the observer and the rest of the universe in all its possible states, including
all feasible variants of compactification and spontaneous symmetry breaking (see Section
10.1). By making a measurement that improves our knowledge of the properties of the
observer, one simultaneously obtains information about the rest of the universe, just as
by measuring the spin of one particle in the Einstein–Podolsky–Rosen experiment one
promptly obtains information about the spin of the other [302, 304, 359].
This answer, in our view, is correct and entirely sufficient. Nevertheless, it would
be more satisfying to have an alternative reply to the foregoing objection, one that is
conceptually simpler and that does not require an analysis of the somewhat obscure
foundations of quantum cosmology for its justification. Moreover, we would like to obtain
an answer to another (and in our opinion, the most important) objection to the Anthropic
Principle, namely that it does not seem at all necessary for the existence of life as we know
nB δρ
it to have identical conditions (homogeneity, isotropy, ratios ∼ 10−9, ∼ 10−5 , etc.)
nγ ρ
over the whole observable part of the universe. The random occurrence of such uniformity
seems completely unlikely.
As noted in Chapter 1, both of these objections can be dispatched particularly sim-
ply by the theory of the self-regenerating inflationary universe. Specifically, long-wave
fluctuations are generated during inflation not only of the inflaton field ϕ, which drives
inflation, but of all other scalar fields Φ with mass mΦ ≪ H as well (and having a small
coupling constant ξ in possible interactions of the type ξ R Φ2 ). In the chaotic inflation
scenario, this means that in certain regions of the universe where the potential energy
density of the scalar field ϕ permanently fluctuates near V(ϕ) ∼ M4P (by virtue of the
self-regeneration process taking place in such regions), long-wave fluctuations of practi-
cally every scalar field Φ grow until the mean value of the potential energy density of
each of these fields is no longer of order M4P . This follows simply from an analysis of the
Hawking–Moss distribution (7.4.1) for the field Φ with V(Φ = 0, ϕ) ∼ M4P .
The upshot of this process is that a distribution of the scalar fields ϕ and Φ is set
up that is quite uniform on exponentially large scales because of inflation; but on the
scale of the universe as a whole, on the other hand, the fields can take on practically any
value for which their potential energy density does not exceed the Planck value. In those
regions of the universe where inflation has ended, the fields ϕ and Φ roll down to different
local minima of V(ϕ, Φ), and since all feasible initial conditions for rolling are realized in
different regions of the universe, the universe becomes divided into different exponentially
large domains containing the fields ϕ and Φ corresponding to all local minima of V(ϕ, Φ),
i.e., all possible types of symmetry breaking in the theory.
At the stage of strong fluctuations with V(ϕ, Φ) ∼ M4P , not only can the magnitudes of
 226

the scalar fields vary, but large metric fluctuations can also be generated, leading to local
compactification or decompactification of space in Kaluza–Klein or superstring theories.
If a region of space with changing compactification is inflating and has an initial size
H−1 ∼ M−1 P (at Planck densities, the probability of this happening should not be too
low), then as a result of inflation this region will turn into an exponentially large domain
with a new type of compactification (for example, a different dimensionality) [78].
Thus, the universe will be partitioned into enormous regions (mini-universes), which
will manifest all possible types of compactification and all possible types of spontaneous
symmetry breaking compatible with the process of inflation, leading to an exponential
growth in the size of these regions. Reference [333] contains an implementation of this
scenario in the context of certain specific Kaluza–Klein theories.
It should be emphasized that due to the unlimited temporal extent of the process
of inflation in a self-regenerating universe, such a universe will support an unbounded
collection of mini-universes of all types, even if the probability of creation of some of
them is extremely small. But this is just what is needed in support of the so-called Weak
Anthropic Principle: we reside in regions with certain space-time and matter properties
not because other regions are impossible, but because regions of the required type exist,
and life as we know it (or, to be more precise, the carbon based life of human observers
of our type) would not be possible in others.5 It is important here that the total volume
of regions in which we could live be unbounded, so that life as we know it will come into
being even if the probability of its spontaneous appearance is vanishingly small. This
does not mean, of course, that we can choose the laws of physics at random. The issue
simply involves choosing one type of compactification and symmetry breaking or another,
as allowed by the theory at hand. The search for theories in which the part of the world
that surrounds us can have the properties that we observe is still a difficult problem, but
it is much easier than a search for theories in which the whole world is not permitted to
have properties different from those in the part where we now live.
Naturally, most of what we have said would remain true if we simply considered an
infinitely large universe with chaotic initial conditions. But when inflation is not taken into
account, the Anthropic Principle is incapable of explaining the uniformity of properties
of the observable part of the universe (see Chapter 1). Furthermore, the mechanism for
self-regeneration of the inflationary universe enables one to put the Anthropic Principle
on a firm footing, given the most natural initial conditions in the universe, and regardless
of whether it is finite or infinite.
We now consider several examples that demonstrate the various ways in which the
Anthropic Principle can be applied in inflationary cosmology.
1) Consider first the symmetry breaking process in supersymmetric SU(5) theory.
After inflation, the universe breaks up into exponentially large domains containing the
fields Φ and H, corresponding to all possible types of symmetry breaking. Among these
domains, there will be some in the SU(5)-symmetric phase and some in the SU(3) ×
5
Zelmanov proposed a similar formulation of the Anthropic Principle [77], saying that we are witnesses
only to certain definite kinds of physical processes because other processes take place without witnesses.
 227

U(1)-symmetric phase, corresponding to the type of symmetry breaking that we observe.

Within each of these domains, the vacuum state will have a lifetime many orders of
magnitude longer than the 1010 years which have passed since the end of inflation in our
part of the universe. We live inside a domain with SU(3) × U(1) symmetry, within which
there are strong, weak, and electromagnetic interactions of the type actually observed.
This occurs not because there are no other regions whose properties differ from ours, and
not because life is totally impossible in other regions, but because life as we know it is
only possible in a region with SU(3) × U(1) symmetry.
2) Consider next the theory of the axion field θ with a potential of the form (7.7.22):
!
θ
V(θ) ∼ m4π 1 − cos √ . (10.5.1)
2 Φ0
√ √
The field θ can take any value in the range − 2 π Φ0 to 2 π Φ0 . A natural estimate for
the initial value of the axion field would therefore be θ = O(Φ0 ), and the initial value of
V(θ) should be of order m4π . An investigation of the rate at which the energy of the axion
12
field falls off as the universe expands shows that for Φ0 > ∼ 10 GeV, most of the energy
density would presently be contributed by axions, while the baryon energy density would
be considerably lower than its presently observed value of ρB > ∼ 2 · 10
−31
g/cm3 . (Since
the universe becomes almost flat after inflation, the overall matter density in the universe
should now be ρc ∼ 2 · 10−29 g/cm3 , independent of the value of the parameter Φ0 .) This
11 12
information was used to derive the strong constraint Φ0 < ∼ 10 –10 GeV [49]. This is not
15 17
a terribly felicitous result, as axion fields with Φ0 ∼ 10 –10 GeV show up in a natural
way in many models based on superstring theory [50].
Let us now take a somewhat closer look at whether one can actually obtain the con-
11 12
straint Φ0 < ∼ 10 –10 GeV in the context of inflationary cosmology.
As we remarked in Section 7.7, long-wave fluctuations of the axion field θ are generated
at the time of inflation (if Peccei–Quinn symmetry breaking, resulting in the potential
(10.5.1), takes place before the end of inflation). By the end of inflation, therefore, a
quasihomogeneous distribution of√the field θ will √ have appeared in the universe, with the
field taking on all values from − 2 π Φ0 to 2 π Φ0 at different points in space with a
probability that is almost independent of θ0 [276, 224]. This means that one can always
find exponentially large regions of space within which θ ≪ Φ0 . The energy of the axion
field always remains relatively low in such regions, and there is no conflict with the
observational data.
12
In and of itself, this fact does not suffice to remove the constraint Φ0 < ∼ 10 GeV, since
when Φ0 ≫ 1012 GeV, only within a very small fraction of the volume of the universe
is the axion field energy density small enough by comparison with the baryon density.
It might therefore seem unlikely that we just so happened make our appearance in one
of these particular regions. Consider, for example, those regions initially containing a
field θ0 ≪ Φ0 , for which the present-day ratio of the energy density of the axion field to
the baryon density is consistent with the observational data. It can be shown that the
total number of baryons in regions with θ ∼ 10 θ0 should be ten times the number in
 228

regions with θ ∼ θ0 . One might therefore expect the probability of randomly winding
up in a region with θ ∼ 10 θ0 (contradicting the observational data) to be ten times that
of winding up in a region with θ ∼ θ0 . Closer examination of this problem indicates,
however, that the mean matter density in galaxies at time t ∼ 1010 years is proportional
to θ8 , and in regions with θ ∼ 10 θ0 it should be 108 times higher than in regions like our
own, with θ ∼ θ0 [334]. A preliminary study of the star formation process in galaxies with
θ ∼ 10 θ0 indicates that solar-type stars are most likely not formed in such galaxies. If
that is really the case, then the conditions required for the appearance of life as we know
it can only be realized when θ ∼ θ0 , and that is exactly why we find ourselves to be in
one such region rather than in a typical region with θ ≫ θ0 . Generally speaking, then,
12
the observational data do not imply that Φ0 < ∼ 10 GeV. In any event, since regions with
12
θ ∼ θ0 most assuredly exist, a derivation of the constraint Φ0 < ∼ 10 GeV would require
that one show that it is much more improbable for life as we know it to arise in regions
with θ ∼ θ0 than in regions with θ ≫ θ0 . As we have already said, an investigation of this
question indicates just the opposite.
The preceding discussions are quite general in nature, and can be applied not just to
the theory of axions, but to the theory of any other light, weakly interacting scalar fields
as well — dilatons, for example [335]. In principle, by applying the Anthropic Principle
to axion cosmology, one might attempt to explain why the present-day baryon density ρB
is 10−1 –10−2 times the total matter density ρ0 ≈ ρc in the universe. Actually, for θ ≪ θ0 ,
the energy density of the axion field would be low (ρa ∼ θ2 ), so the main contribution
to ρ0 would come from baryons, ρ0 ≈ ρc ≈ ρB . But only a small fraction of the baryons
θ
in the universe (proportional to ) are to be found in regions with θ ≪ θ0 . At the
Φ0
same time, for θ ≫ θ0 , the conditions of life would be markedly different from our own,
and it is most unlikely that one would find himself inside such a region of the universe.
The location of the maximum probability for the existence of life as we know it, as a
function of θ, depends on the value of Φ0 , and for a particular value Φ0 ≫ 1012 GeV, the
maximum may be attained precisely in a state with initial value θ ∼ θ0 , in which now
ρB ∼ 10−1–10−2 ρ0 . Consequently, a study of the theory of star and galaxy formation
together with a detailed study of the conditions necessary for the existence of life as we
know it may actually enable us to determine the most likely value of the parameter Φ0 in
the theory of axions.
3) The preceding results can be employed practically unchanged to avoid one of the
principal difficulties encountered in using the mechanism for generating the baryon asym-
metry of the universe proposed by Affleck and Dine [97, 98]. Recall that the baryon
nB
asymmetry produced by this mechanism is typically too large: the value of ranges
nγ
from −O(1) to +O(1), depending on the magnitude of the angle θ in isotopic spin space
between the initial values of two different scalar fields. According to the inflationary
universe scenario, one can always find exponentially large regions in which this angle is
nB
small and ∼ 10−9. Such regions occupy a very small fraction of the total volume
nγ
 229

nB
of the universe. But in regions, say, with ∼ 10−7 , the density of matter in galaxies
nγ
will be eight orders of magnitude greater than in our own, and life as we know it will
either be impossible or extremely unlikely. Naturally, there are a number of other ways
to get rid of the excess baryon asymmetry of the universe (see Section 7.10), but inter-
estingly enough, the Anthropic Principle as applied within the scope of the theory of the
inflationary universe may turn out to be sufficient to resolve this problem all by itself.
4) The last example that we present here is somewhat different from its predecessors.
We know that in the standard Friedmann cosmology, the universe, if it is closed, will
spend approximately half its life in a state of expansion, and the other half in a state of
contraction. A similar phenomenon can also occur locally in an inflationary universe on
scales at which density inhomogeneities that came into being during the inflationary stage
δρ
become large, with ∼ 1 [336]. The question that arises is then ‘Why is the observable
ρ
part of the universe expanding?’ Do we live in an expanding part of the universe by
accident, or is there some special reason for this circumstance?
λ
The answer to this question is related to the fact that in the simplest ϕ4 theory with
4
λ ∼ 10−14 , for example, the size of a homogeneous locally Friedmann part of the universe
6·104
is of order l ∼ M−1P exp(π λ
−1/3
) ∼ M−1P · 10 (1.8.8), and a typical mass concentrated
2·105
in such a region is of order M ∼ MP · 10 , so according to (1.3.15), the typical time
5
until the beginning of contraction within such a region is t ∼ 102·10 years [336]. In a
self-regenerating universe, inasmuch as it exists without end, there should be regions that
are much older and regions that are much younger. We happen to live in a relatively
young region, which has existed a total of 1010 years since the end of inflation (in this
region). This is related to the fact that life as we know it exists near solar-type stars,
whose maximum lifetime is 1010 –1011 years. This is precisely why the part of the universe
that surrounds us is still in the initial stage of its expansion, and that expansion (within
5
the framework of the simple model considered here) should last at least another 102·10
years.
Taken by itself, the foregoing certainly does not mean that no life at all is possible
during the contraction stage [336]; the issue is simply that at the current rate of evolution
of living organisms (and also taking into account the probable decay of baryons after
5
1035 –1040 years), observers 102·10 years hence will scarcely resemble the way we are now.
We wish to emphasize once again that the so-called Weak Anthropic Principle, as
formulated and used above, is conceptually quite simple. It involves an assessment of
the probability of observing a region of the universe with given properties, under the
condition that the fundamental properties of the observer himself also be known. The
preceding discussions require no philosophical sophistication, and their import is of a
trivial, mundane nature — for instance, we live on the surface of the Earth not because
there is more room here than in interstellar space, but simply because in interstellar space
we could never breathe.
Furthermore, the richness and heuristic value of the results obtained via the Weak
Anthropic Principle have impelled many authors to try to expand and generalize it as much
 230

as possible, even if such a generalization is presently not entirely justified (see [331]). The
possibility of such a generalization is suggested by the unusually important role played by
the concept of the observer in the construction and interpretation of quantum cosmology.
Most of the time, when discussing quantum cosmology, one can remain entirely within the
bounds set by purely physical categories, regarding the observer simply as an automaton,
and not dealing with questions of whether he has consciousness or feels anything during
the process of observation [305]. This we have done in all of the preceding discussions.
But we cannot rule out the possibility a priori that carefully avoiding the concept of
consciousness in quantum cosmology constitutes an artificial narrowing of one’s outlook.
A number of authors have underscored the complexity of the situation, replacing the
word observer with the word participant, and introducing such terms as “self-observing
universe” (see, for example, [302, 323]). In fact, the question may come down to whether
standard physical theory is actually a closed system with regard to its description of the
universe as a whole at the quantum level: is it really possible to fully understand what
the universe is without first understanding what life is?
Leaving aside the question of how well motivated such a statement of the problem
is, let us note that similar problems often appear in the development of science. We
know, for example, that classical electrodynamics is incomplete, an example being the
problem of the self-acceleration of an electron, requiring the use of quantum theory for a
solution [65]. Quantum electrodynamics likely suffers from the zero-charge problem [156,
157], which can be circumvented by including electrodynamics in a unified nonabelian
gauge theory [3]. The quantum theory of gravitation presents even greater conceptual
difficulties, and attempts have been made to overcome these by significantly broadening
and generalizing the original theory [14–17]. We do not know whether it is possible
to assign an exact meaning to many of the concepts employed in quantum cosmology
(probability of creation of the universe from “nothing,” splitting of the universe in the
many-worlds interpretation of quantum mechanics, etc.) without stepping outside the
confines of the existing theory, and possible ways of generalizing this theory are still far
from clear. The only thing we can do at this point is to attempt to draw upon analogies
from the history of science which may prove to be instructive.
Prior to the advent of the special theory of relativity, space, time, and matter seemed
to be three fundamentally different entities. In fact, space was thought to be a kind of
three-dimensional coordinate grid which, when supplemented by clocks, could be used to
describe the motion of matter. Special relativity did away with the insuperable distinction
between space and time, combining them into a unified whole. But space-time nevertheless
remained something of a fixed arena in which the properties of matter became manifest.
As before, space itself possessed no intrinsic degrees of freedom, and it continued to play
a secondary, subservient role as a backdrop for the description of the truly substantial
material world.
The general theory of relativity brought with it a decisive change in this point of view.
Space-time and matter were found to be interdependent, and there was no longer any
question of which was the more fundamental of the two. Space-time was also found to
have its own inherent degrees of freedom, associated with perturbations of the metric
 231

— gravitational waves. Thus, space can exist and change with time in the absence of
electrons, protons, photons, etc.; in other words, in the absence of anything that had pre-
viously (i.e., prior to general relativity) been subsumed by the term matter. (Note that
because of the weakness with which they interact, gravitational waves are exceedingly
difficult to detect experimentally, an as-yet unsolved problem.)
A more recent trend, finally, has been toward a unified geometric theory of all funda-
mental interactions, including gravitation. Prior to the end of the 1970’s, such a program
— a dream of Einstein’s — seemed unrealizable; rigorous theorems were proven on the
impossibility of unifying spatial symmetries with the internal symmetries of elementary
particle theory [337]. Fortunately, these theorems were sidestepped after the discovery of
supersymmetric theories [85]. In principle, with the help of supergravity, Kaluza–Klein,
and superstring theories, one may hope to construct a theory in which all matter fields
will be interpreted in terms of the geometric properties of some multidimensional super-
space [13–17]. Space would then cease to be simply a requisite mathematical adjunct for
the description of the real world, and would instead take on greater and greater inde-
pendent significance, gradually encompassing all the material particles under the guise
of its own intrinsic degrees of freedom. Of course, this does not at all mean that the
concept of matter becomes useless. The issue at hand is simply the revelation of the
fundamental unity of space, time, and matter, which is hidden from us in much the
same way that the unity of the weak and electromagnetic interactions was hidden until
recently.
According to standard materialistic doctrine, consciousness, like
space-time before the invention of general relativity, plays a secondary, subservient role,
being considered just a function of matter and a tool for the description of the truly ex-
isting material world. It is certainly possible that nothing similar to the modification and
generalization of the concept of space-time will occur with the concept of consciousness in
the coming decades. But the thrust of research in quantum cosmology has taught us that
the mere statement of a problem which might at first glance seem entirely metaphysical
can sometimes, upon further reflection, take on real meaning and become highly signif-
icant for the further development of science. We should therefore like to take a certain
risk and formulate several questions to which we do not yet have the answers.
Is it not possible that consciousness, like space-time, has its own intrinsic degrees
of freedom, and that neglecting these will lead to a description of the universe that is
fundamentally incomplete? Will it not turn out, with the further development of sci-
ence, that the study of the universe and the study of consciousness will be inseparably
linked, and that ultimate progress in the one will be impossible without progress in the
other? After the development of a unified geometrical description of the weak, strong,
electromagnetic, and gravitational interactions, will the next important step not be the
development of a unified approach to our entire world, including the world of conscious-
ness?
All of these questions might seem somewhat naive and out of place in a serious scientific
publication, but to work in the field of quantum cosmology without an answer to these,
and without even trying to discuss them, gradually becomes as difficult as working on
 232

the hot universe theory without knowing why there are so many different things in the
universe, why nobody has ever seen parallel lines intersect, why the universe is almost
homogeneous and looks approximately the same at different locations, why space-time
is four-dimensional, and so on (see Section 1.5). Now, with plausible answers to these
questions in hand, one can only be surprised that prior to the 1980’s, it was sometimes
taken to be bad form even to discuss them. The reason is really very simple: by asking
such questions, one confesses one’s own ignorance of the simplest facts of daily life, and
moreover encroaches upon a realm which may seem not to belong to the world of positive
knowledge. It is much easier to convince oneself that such questions do not exist, that
they are somehow not legitimate, or that someone answered them long ago.
It would probably be best then not to repeat old mistakes, but instead to forthrightly
acknowledge that the problem of consciousness and the related problem of human life and
death are not only unsolved, but at a fundamental level they are virtually completely
unexamined. It is tempting to seek connections and analogies of some kind, even if they
are shallow and superficial ones at first, in studying one more great problem — that of
the birth, life, and death of the universe. It may conceivably become clear at some future
time that these two problems are not so disparate as they might seem.

10.6 Quantum cosmology and the signature of space-time

The most significant modification to the concept of four-dimensional space-time that we
have discussed so far is that of a space with one temporal and d − 1 spatial coordinates,
some of the latter being compactified. This construction, however, is clearly not the most
general. Our intuitive ideas about space-time are linked to our study of the dynamics
of objects whose dimensions may be arbitrarily small, but in the quantum theory of
gravitation it is difficult (or impossible) to consider objects smaller than M−1 P . If the
theory is to be based on the study of extended objects like strings or membranes, many
of our intuitive ideas about the geometrical objects associated with them (points, straight
lines, etc.) will be found to be largely inadequate [17].
Unanswered questions arise, however, even at a simpler level. For example, why are
there many spatial coordinates, but only one temporal coordinate? That is, why does our
space have the signature (+ − − − . . . −)? Why could it not be Euclidean, with signature
(+ + . . . +) or (− − . . . −)? Why are just the spatial dimensions compactified, and not the
temporal? Are transformations which change the signature of the metric possible [292]?
In the context of a model of the universe consisting of large regions with differing
properties, these may all prove to be sensible questions. It is therefore worth considering,
if only briefly, how the properties of the universe would be altered under a change in the
signature of the metric. There are many aspects to this question, some of which stand out
particularly clearly in supergravity and the theory of superstrings. For instance, the 16-
component Majorana–Weyl spinors required for a formulation of supergravity in a space
with d = 10 only exist for three different signatures of the metric: 1 + 9 (one temporal
 233

and nine spatial dimensions), 5 + 5, and 9 + 1 [338]; a supersymmetric theory has only
been formulated in the first case.
There exists one additional more general problem that arises in a very broad class of
theories when space contains more than one time dimension. The problem is most readily
understood by studying a scalar field in a flat space with signature (+ + −−) as an
example. The usual dispersion relation for the field ϕ, which takes the form k02 = k2 + m2
in Minkowski space, then becomes

k02 = k22 + k32 + m2 − k12 . (10.6.1)

Here the momentum k1 can clearly change the sign of the effective mass squared in (10.6.1),
or in other words, it can induce exponentially rapid growth of fluctuations of the field ϕ
when k12 > k22 + k32 + m2 :
q
δϕ ∼ exp( k12 − k22 − k32 − m2 t) . (10.6.2)

This effect is analogous to the instability of the vacuum state with ϕ = 0 in the theory
of a scalar field with negative mass squared (see (1.1.5), (1.1.6)). In the theory (1.1.5),
however, the development of the instability came to a halt when the sign of the effec-
tive squared mass m2 (ϕ) changed as the field ϕ increased. But in the present example,
the instability grows without bound, since there are exponentially growing modes with
arbitrary values of m2 for sufficiently large momenta k1 . Since the instability is asso-
ciated precisely with the very largest momenta (shortest wavelengths), the existence of
such an instability will most likely be a general feature of theories in a space with more
than one time dimension, regardless of either the topology of the space or whether the
additional time dimensions are compactified. In some theories it proves to be possible to
avoid instability in modes corresponding to particles that have relatively low mass after
compactification [293], but there still remains an instability due to heavy particles with
masses m of the order of the reciprocal of the compactification radius R−1 c . From (10.6.2),
it follows that this instability is not the least bit less dangerous. One might hold out hope
that for some reason there might be a cutoff at k0 , k1 ∼ R−1 c in this theory; modes with
>
m ∼ Rc might then not appear. But if there were a cutoff at momenta of order R−1
−1
c , it
would become impossible even to discuss compactification in conventional semiclassical
language. To put it differently, until such time as we are capable of describing a classical
space containing more than one time dimension, instability seems unavoidable.
In Euclidean space there is no instability, but there is also no evolution over the course
of time, which is necessary for the existence of life as we know it. Furthermore, Euclidean
space also lacks the requisite instability with respect to exponential growth of the universe,
which leads to inflation and makes the universe so large.
To summarize this section, we might say that where there is no time, there is neither
evolution nor life, and where there is too much time, instability is rampant and life is
short. From this perspective, the familiar signature of the metric seems to be a necessary
condition for progress to take place within a relatively orderly framework.
 234

10.7 The cosmological constant, the Anthropic Principle, and reduplication

of the universe and life after inflation
As noted in Section 1.5, one of the most difficult problems in modern physics is the
problem of the vacuum energy, or the cosmological constant. There have been a great
many interesting suggestions as to how this problem might be solved — for example, see
[17, 78, 116, 292, 335, 339–359]. This multitude of proposals can be divided into two
main groups. The most attractive possibility is that due to some mechanism related to
a symmetry of the theory, for example, the vacuum energy must be exactly zero. The
second possibility, presently an active topic of discussion among experts in the theory
of the formation of the large-scale structure of the universe, is that there is some sort
of mechanism that may engender a vacuum energy density ρν at the present time of
the same order of magnitude as the present total matter density ρ0 ∼ ρc ∼ 2 · 10−29
g/cm3 . While deep-seated reasons may exist for the vanishing of the vacuum energy,
however, ensuring the equality of ρν and ρ0 at the present epoch (if only to order of
magnitude) is difficult without placing unnatural constraints on the parameters of the
theory.
The Anthropic Principle suggests one possible escape from this situation. To illus-
trate the basic idea behind this approach to the problem of the cosmological constant,
consider the theory of a scalar field Φ with effective potential V(Φ, ϕ) = α M3P Φ + V(ϕ)
[78, 341]. Here V(ϕ) is the potential of the field ϕ responsible for inflation, with a min-
imum at the point ϕ0 . We shall assume that the constant α is very small, α < ∼ 10
−120
.
Fluctuations of the field Φ that set in at the time of inflation result in space being par-
titioned into regions with all possible values of V(Φ, ϕ0 ), ranging from −M4P to +M4P .
In those regions where V(Φ, ϕ0 ) ≪ −10−29 g/cm3 at present, the universe looks locally
like de Sitter space with negative vacuum energy. In such regions, all structures come
into being and pass out of existence in much less than 1010 years, and life as we know
it cannot emerge. In regions with V(Φ, ϕ) > 2 · 10−29 g/cm3 , inflation is still ongo-
ing, and if the potential V(Φ, ϕ0 ) is very flat (α <∼ 10
−120
) the field Φ will vary quite
slowly; the time needed for V(Φ, ϕ0 ) to decrease to 10−29 g/cm3 will then be more than
1010 years. In regions with V(Φ, ϕ0 ) >∼ 10
−27
g/cm3 , the standard mechanism for galaxy
formation is altered significantly, and for V(Φ, ϕ0 ) ≫ 10−27 g/cm3 , galaxies and stars
like our own are hardly formed at all [348]. This still does not tell us why presently
V(Φ, ϕ0 ) <
∼ 10
−29
g/cm3 , but the fact that the observational constraints on the vacuum
energy density must be satisfied in at least a few percent of the “habitable” reaches of
the universe makes the problem of the cosmological constant much less acute. An even
better model would be one in which the spectrum of possible values of the vacuum en-
ergy ρv is discrete rather than continuous, and includes states with ρv = 0 but not those
with energy density less than 10−27 g/cm3 . With the enormous number of possible types
of compactification in Kaluza–Klein theories, such a possibility can actually be realized
[292], and a similar possibility emerges in superstring theory [353]. In any case, the very
fact that the Anthropic Principle may enable us to narrow the range of possible values
of ρv in the observable part of the universe from −1094 g/cm3 < ∼ ρv <
94
∼ 10 g/cm to
3

−10−29 g/cm3 < ∼ ρv <∼ 10

−27
g/cm3 (i.e., to reduce the range by a factor of 10121 ) seems
 235

worthy of note.
We will soon return to a discussion of using the Anthropic Principle to solve the prob-
lem of the cosmological constant. Here we consider the possibility that the cosmological
constant vanishes because of some hidden symmetry. Several suggestions have been made
on this score. One of the most interesting and promising ideas has to do with the appli-
cation of supersymmetric theories, and superstring theories in particular [17]. In certain
versions of such theories, the vacuum energy in the absence of supersymmetry breaking
vanishes to all orders of perturbation theory [17, 353]. In the real world, however, super-
symmetry is broken, and it is still unclear whether the vacuum energy remains at zero
after supersymmetry breaking. Another possibility has to do with so-called dilatation
invariance, which (if certain rather strong assumptions are made) might be of some help
despite the fact that it is also broken [335] (see also [354]). Below, we discuss one more
possibility with a direct bearing on quantum cosmology that shows how many surprises
this science may hold in store [344].
Consider a model that simultaneously describes two different universes X and X̄, with
coordinates xµ and x̄α respectively (µ, α = 0, 1, 2, 3) and metrics gµν (x) and ḡαβ (x̄),
containing the fields ϕ(x) and ϕ̄(x̄) with action of the following unusual type [344]:6
Z q q
S = N d4 x d4 x̄ g(x) ḡ(x̄)
M2P M2
( )
× R(x) + L[ϕ(x)] − P R(x̄) − L[ϕ̄(x̄)] . (10.7.1)
16 π 16 π

Here N is a normalizing constant. The action (10.7.1) is invariant under general covariant
transformations in each of the universes individually. A novel symmetry of the action
(10.7.1) is the symmetry under the transformations ϕ(x) → ϕ̄(x), gµν (x) → ḡαβ (x),
ϕ̄(x̄) → ϕ(x̄), ḡαβ (x̄) → gµν (x̄) with a subsequent change of sign, S → −S. For reasons
that will soon become clear, we call this antipodal symmetry. (In principle, other terms
that do not violate this symmetry could be added to the integrand of Eq. (10.7.1), such
as any odd function of ϕ(x) − ϕ̄(x̄); this would not affect our basic result.)
One immediate consequence of antipodal symmetry is invariance under a shift in the
values of the effective potentials V(ϕ) → V(ϕ) + C, V(ϕ̄) → V(ϕ̄) + C, where C is an
arbitrary constant. Thus, nothing in the theory depends on the value of the potentials
V(ϕ) and V(ϕ̄) at their absolute minima ϕ0 and ϕ̄0 (note that ϕ0 = ϕ̄0 and V(ϕ0 ) = V(ϕ̄0 )
by virtue of the same symmetry). This is precisely why it proves possible to solve the
cosmological constant problem in the theory (10.7.1).
However, the main reason for invoking this new symmetry was not just to solve the
cosmological constant problem. Just as the theory of mirror particles was originally
proposed in order to make the theory CP-symmetric while maintaining CP-asymmetry
in its observable sector, the theory (10.7.1) is proposed in order to make the theory
symmetric with respect to the choice of the sign of energy. This removes the old prejudice
that even though the overall change of sign of the Lagrangian (i.e., of both its kinetic
6
Somewhat different (but similar) models have also been considered elsewhere [116, 293].
 236

and potential terms) does not change the solutions of the theory, one must say that the
energy of all particles is positive. This prejudice was so strong that several years ago people
preferred to quantize particles with negative energy as antiparticles with positive energy,
which resulted in the appearance of such meaningless concepts as negative probability.
We wish to emphasize that there is no problem with performing a consistent quantization
of theories which describe particles with negative energy. Difficulties appear only when
there exist interacting species with both signs of energy. (As noted in Section 10.1, this
is one of the main problems of quantum cosmology, where one must quantize fields with
positive energy and the scalar factor a with negative energy.) In the present case no
such problem exists, just as there is no problem of antipodes falling off the opposite side
of the Earth. The reason is that the fields ϕ̄(x̄) do not interact with the fields ϕ(x),
and the equations of motion for the fields ϕ̄(x̄) are the same as for the fields ϕ(x) (the
overall minus sign in front of L[ϕ̄(x̄)] does not change the Lagrange equations). In other
words, in spite of the fact that from the standpoint of the sign of the energy of matter,
universe X̄ is an antipodal world where everything is upside-down, there is no instability
there, and particles of the field ϕ̄(x̄) are completely unaware that they have energy of the
“wrong” sign, just as our antipodal counterparts living on the other side of the globe are
completely unruffled by the fact that they are upside-down from our point of view.
Similarly, gravitons from different universes do not interact with each other. However,
some interaction between the two universes does exist. In the theory (10.7.1), the Einstein
equations take the form
1
Rµν (x) − gµν (x) R(x) = −8 π G Tµν (x)
2 * +
R(x̄)
− gµν (x) + 8 π G L[ϕ̄(x̄)] ,
2
(10.7.2)
1
Rαβ (x̄) − gαβ (x̄) R(x̄) = −8 π G Tαβ (x̄)
2 * +
R(x)
− gαβ (x̄) + 8 π G L[ϕ(x)] .
2
(10.7.3)

Here G = M−2
P , Tµν is the energy-momentum tensor of the field ϕ(x), Tαβ is the energy-
momentum tensor of the field ϕ̄(x̄), and the meaning of the angle-bracket notation is
Z q
d4 x g(x) R(x)
hR(x)i = Z q , (10.7.4)
4
dx g(x)
Z q
d4 x̄ ḡ(x̄) R(x̄)
hR(x̄)i = Z q , (10.7.5)
d4 x̄ ḡ(x̄)
 237

and similarly for hL[ϕ(x)]i and hL[ϕ̄(x̄)]i. Thus, although particles in universes X and X̄
do not interact with each other, the universes themselves do interact, but only globally:
each one makes a time-independent contribution to the average vacuum energy density
of the other, with averaging taking place over the entire history of the universe. At
the quantum cosmological level, when one writes down the equations for universe X, for
example, averaging should take place over all possible states of universe X̄ — that is, the
result should not depend on the initial conditions in each of the two universes.
Generally speaking, it is extremely difficult to calculate the averages (10.7.4) and
(10.7.5), but in the inflationary universe scenario (at least at the classical level), everything
turns out to be quite simple. Indeed, after inflation the universe becomes almost flat, and
its lifetime becomes exponentially long (or even infinite, if it is open or flat). In that
event, the dominant contribution to the mean values hRi and hLi comes from the late
stages of the evolution of the universe, when the fields ϕ(x) and ϕ̄(x̄) relax near the global
minima of V(ϕ) and V(ϕ̄). As a consequence, the mean value of −L[ϕ(x)] is the same, to
exponentially high accuracy, as the value of the potential V(ϕ) at its global minimum at
ϕ = ϕ0 , and the mean value of the curvature scalar R(x) is identical to its value during
the late stages of evolution of the universe X, when the universe transforms to the state
ϕ = ϕ0 , corresponding to the global minimum of V(ϕ). The analogous statement also
holds true for hL[ϕ̄(x̄)]i and hR(x̄)i. For that reason, Eqs. (10.7.2) and (10.7.3) take the
following form in the late stages of evolution of the universes X and X̄:
1
Rµν (x) − gµν (x) R(x) = 8 π G gµν (x) [V(ϕ̄0 ) − V(ϕ0 )]
2
1
− gµν (x) R(x̄) , (10.7.6)
2
1
Rαβ (x̄) − gαβ (x̄) R(x̄) = 8 π G gαβ (x̄) [V(ϕ0 ) − V(ϕ̄0 )]
2
1
− gαβ (x̄) R(x) , (10.7.7)
2
yielding

R(x) = 2 R(x̄) + 32 π G [V(ϕ0 ) − V(ϕ̄0 )] , (10.7.8)

R(x̄) = 2 R(x) + 32 π G [V(ϕ̄0 ) − V(ϕ0 )] . (10.7.9)

Recall from our earlier discussion that ϕ0 = ϕ̄0 and V(ϕ0 ) = V(ϕ̄0 ) by virtue of antipodal
symmetry. This implies that in the late stages of evolution of the universe X,
32
R(x) = −R(x̄) = G [V(ϕ0 ) − V(ϕ̄0 )] = 0 . (10.7.10)
3
We emphasize that the contribution made by universe X̄ to the effective vacuum energy
of universe X does not depend on the time t in the latter. The cancellation represented
by (10.7.10) therefore takes place only in the late stages of evolution of universe X, and
its sole effect is to add a constant term to V(ϕ) in such a way as to obtain V(ϕ0 ) = 0.
 238

Thus, the mechanism considered here does not alter the standard inflationary scenario at
all.
Note that this model differs from the conventional Kaluza–Klein theory in which, as
we have already stated, the introduction of two time coordinates immediately leads to
instability. If would be straightforward to generalize the theory (10.7.1), for example, by
writing the action as an integral over universes X1 , X2 , . . ., and taking the Lagrangian to
be a sum of Lagrangians from the various fields ϕ1 (x1 ), ϕ2 (x2 ), . . ., each of which resides
in only one of these universes. With such a scheme, our world would consist of arbitrarily
many different universes interacting with each other only globally, the inhabitants of each
having their own time and their own physical laws. This approach would provide a basis
for the Anthropic Principle in its strongest form.
Of course there are shortcomings to be dealt with. This scheme could be generalized
to supersymmetric theories, but it would be difficult to do so in a way that ensures that
the cosmological constant vanishes automatically. If the universe is self-regenerating, one
could encounter difficulties in calculating the averages (10.7.4), (10.7.5), since they may
become infrared-divergent and the result may depend on the cutoff. This question has
yet to be thoroughly examined, due to the very complicated large-scale structure of the
self-regenerating universe. However, one can easily avoid such questions in theories in
which V(ϕ) grows rapidly enough at ϕ ≥ ϕ∗ , since there will be no self-regeneration of
the universe in such theories. Another problem is that the integral over d4 x̄ in (10.7.1)
renormalizes the effective Planck constant in the universe X, and one should take a very
small normalization factor (N ∼ exp(−λ1/3 ) in the λ ϕ4 theory) in order to compensate
this renormalization. A further possibility is that in constructing the quantum theory
in a reduplicated universe, one should only do so in each of the noninteracting universes
individually, without taking the foregoing renormalization of N into consideration. Note
also that the mechanism for cancellation of the cosmological term that was suggested
above works independently of the value of N.
In any case, the very fact that (at least at the classical level) there is a large class of
models within which the cosmological constant automatically vanishes, regardless of the
detailed structure of the theory, seems noteworthy. Moreover, the possibility of building
a consistent theory of many universes that interact with one another only globally may
be of interest in its own right.
A very interesting and nontrivial generalization of the ideas we have discussed here has
been proposed quite recently in papers by Coleman [345, 346], Giddings and Strominger
[349], and Banks [347]; these are based on earlier work by Hawking [350], Lavrelashvili,
Rubakov, and Tinyakov [351], and Giddings and Strominger [352] dealing with wormholes
and the loss of coherence in quantum gravitation, as well as on a paper by Hawking [340]
that treated a possible mechanism for the vanishing of the cosmological constant in the
context of quantum cosmology.
The basic idea in Refs. 345–347 and 349 is that because of quantum effects, the
universe can split into several topologically disjoint but globally interacting parts. Such
processes can take place at any location in our universe (see [350–352], [133], and Section
10.3). The baby universes can carry off electron-positron pairs or some other combination
 239

of particles and fields if they are not prevented from doing so by conservation laws. The
simplest way to describe this effect is to say that the existence of baby universes leads to
a modification of the effective Hamiltonian that describes particles and fields in our own
universe [345, 349]:
X
H(x) = H0 (ϕ(x), ψ(x), . . .) + Hi (ϕ(x), ψ(x), . . .) Ai . (10.7.11)

This Hamiltonian describes the fields ϕ, ψ, . . . in our universe on scales exceeding M−1 P . In
(10.7.11), H0 is the part of the Hamiltonian unrelated to topological fluctuations; the Hi
are certain local functions of the fields ϕ, ψ, . . .; the Ai represent a combination of creation
and annihilation operators in the baby universes. Thus, for example, a term like H1 A1 ,
where H1 is constant, is associated with the possibility of a change in the vacuum energy
density due to an interaction with baby universes, the term ē(x) e(x) A2 is associated with
possible electron-positron pair exchange, and so on. The operators Ai are independent of
the position x, since baby universes cannot carry off energy or momentum. According to
[345, 346], the requirement that the Hamiltonian in our universe be local,

[H(x), H(y)] = 0 (10.7.12)

for spacelike x − y, implies that all of the operators Ai commute with each other. They
can therefore all be simultaneously diagonalized by the “α-states” |αi i, so that

Ai |αi i = αi |αi i . (10.7.13)

If the quantum state of the universe is an eigenstate of the operators Ai , then one con-
sequence of the complicated structure of the vacuum (10.7.13) will be the introduction
of an infinitude of a priori undetermined parameters into the effective Hamiltonian: in
(10.7.11), one simply replaces the operators Ai by their eigenvalues αi in the given state.
If the universe is not originally in an eigenstate of the Ai , its wave function, after a series
of measurements, will nonetheless quickly be reduced to such an eigenstate [345].
This makes it possible to consider anew many of the fundamental problems of physics.
It is often supposed that the basic goal of theoretical physics is to find exactly what
Lagrangian or Hamiltonian correctly describes our entire world. But one could well ask
the following question: if we assume that there was a time when our universe (or the
part in which we live) did not exist (at least as a classical space-time), in what sense
can one speak of the existence of laws at that time which would have governed its birth
and evolution? We know, for example, that the laws that control biological evolution are
recorded in our genetic code. But where were the laws of physics recorded if there was no
universe?
One possible answer is that the final structure of the effective Hamiltonian, including
whatever values are taken on by the constants αi , is not fixed until a series of measurements
have been made, determining with finite accuracy which of the possible quantum states
of the universe |αii we live in [355]. This implies that the concept of an observer may play
an important role not just in discussions of the various characteristics of our universe, but
in the very laws by which it is governed.
 240

In general, the wave function of the universe can depend on the parameters αi . This
possibility lies at the root of Coleman’s explanation for the vanishing of the cosmological
constant
8 π V(ϕ0 )
Λ= ,
M2P
where V(ϕ0 ) is the present value of the vacuum energy density. The basic idea goes
back to a paper by Hawking [340], who made use of the Hartle–Hawking wave function
(10.1.12), (10.1.17). According to Hawking, if the cosmological constant could for some
reason take on arbitrary values, the probability of winding up in a universe with a given
value of Λ is
3 M4P 3 π M2P
P(Λ) ∼ exp[−SE (Λ)] = exp = exp (10.7.14)
8V Λ
(compare with (10.1.18)). In the context of the approach that we are considering, which
is based on the theory (10.7.11), (10.7.13), the cosmological constant, like any other
constant, could actually take on different values, depending on exactly what quantum
state we happen to be in. In calculating P(Λ) for this case, however, one must also sum
over all topologically disconnected configurations of the baby universes, which leads to a
modified expression for P(Λ) [346]:

3 π M2P
!
P(Λ) ∼ exp exp . (10.7.15)
Λ

From (10.7.14) and (10.7.15), we see that P(Λ) is sharply peaked at Λ = 0, i.e. among
all possible universes, the most probable are those with a vanishingly small cosmological
constant.
Just how reliable is this conclusion? At the moment, it is difficult to say. In fact,
the probability of formation of Baby universes, resulting in a gravitational vacuum with
highly complicated structure, has yet to be firmly established. The description of this
process using Euclidean methods [350–352] differs from that obtained via the stochastic
approach [133] (see Section 10.3, Eq. (10.3.7)). Furthermore, as noted in Section 10.2,
the use of the Hartle–Hawking wave function in the inflationary cosmology is justified
only when there exists a stationary distribution of the field ϕ, and therefore of V(ϕ) and
Λ(ϕ). Thus far, it has not been possible to find any inflationary models in which such a
stationary distribution might actually exist. The probability distribution for the quantity
Λ(αi ) ought to have been stationary, but this is the probability distribution for finding a
cosmological constant equal to Λ in different universes (or to be more precise, in different
quantum states of the universe), rather than in different parts of a single universe. The
stochastic approach is not capable of justifying expressions like (10.7.14) and (10.7.15)
under these circumstances, and validity of Euclidean methods used for the derivation of
Eq. (10.7.15) is not quite clear.
Some authors argued that a correct distribution of probability for the universe to be in
a given quantum state |αii does not have a peak at Λ = 0, in contrast with Eq. (10.7.15)
[356, 357]. It may be important also that actually we are not asking why the universe lives
 241

in a given quantum state |αi i. We are just trying to explain the experimental fact that
we live in the universe in the quantum state |αi i corresponding to ρv = |V(ϕ0 )| <
∼ 10
−29

g/cm3 [358, 359].

In this regard, recall that the application of the Anthropic Principle based on an
analysis of the galaxy formation process enables one to place constraints on the vacuum
energy density [348],

−10−29 g/cm3 <

∼ V(ϕ0 ) <
∼ 10
−27
g/cm3 ,

and these constraints are quite close to the experimental figure,

−29
|V(ϕ0 )| <
∼ 10 g/cm3 .

This is an especially interesting result when viewed in the context of the baby universe
theory which makes it possible to choose among the various Λ [345].
Can the Anthropic constraint on the vacuum energy density be made stronger, so as
to make the inequality V(ϕ0 ) < ∼ 10
−29
g/cm3 a necessary consequence of the Anthropic
Principle? There is still no final answer, but there are a few inklings of how one might go
about solving this problem.
As follows from the results obtained in Sections 10.2–10.4, life in all its possible forms
will appear again and again in different domains of the self-regenerating inflationary uni-
verse. This does not mean, however, that one can be very optimistic about the fate of
mankind. An investigation of this question reveals that within the presently observable
part of the universe, life as we know it cannot endure indefinitely, due to the decay of
baryons and the local collapse of matter [336]. The only possibility that we are presently
aware of for the perpetual promulgation of life is that in the scenario under consideration,
λ ϕ4
for example in the theory, there must now exist a large number of domains in every
4
region of size

π (ϕ∗ )2
!
∗ 30
l>
∼l ∼ 10 M−1
P exp
M2P
π (ϕ∗ )2
!
∼ 1030 M−1
P exp (10.7.16)
M2P

in which the process of inflation continues unabated, and will do so forever. There will
always be sufficiently dense regions (like our own) near such domains, in which inflation
came to an end relatively recently and in which baryons have not yet had a chance to
decay. One possible survival strategy for mankind might consist of continual spaceflight
bound for such regions. In the worst case, if we will be unable to travel to such distant
places ourselves, we can try to send some information about us, our life and our knowledge,
and maybe even stimulate development of such kinks of life there, which would be able to
receive and use this information. In such a case one would have a comforting thought that
even though life in our part of the universe will disappear, we will have some inheritors,
 242

and in this sense our existence is not entirely meaningless. (At least it would be not worse
than what we have here now.)
Leaving aside the question of the optimal strategy for the survival of mankind, we
wish to note that the appropriate process is necessarily impossible if the vacuum energy
density V(ϕ0 ) is greater than

V∗ ∼ ρ0 · 10200 exp(−6 π λ−1/3 ) . (10.7.17)

In the theory with λ ∼ 10−14 , V∗ is vanishingly small:

6
V∗ ∼ 10−5·10 g/cm3 ≪ ρ0 . (10.7.18)

The reason there is a critical value V∗ is that when V(ϕ0 ) > V∗ , the size H−1 (ϕ0 ) of
the event horizon in a world with vacuum energy density V(ϕ0 ) turns out to be less
than the typical distance between domains in which the process of self-regeneration of
the universe is taking place. (This distance is presently l∗ of (10.7.16); by the time the
vacuum energy V(ϕ0 ) begins to dominate, the distance will have grown by a factor of
approximately 10−60 exp(2 π λ−1/3 ).) Under such circumstances, it would be impossible
in principle either to fly or to send a signal from our region of the universe to regions in
the vicinity of self-regenerating domains; see Section 1.4.
Thus, in the present model, any quantum state of the universe |αi i with vacuum energy
−5·106
density V(ϕ0 ) > ∼ 10 g/cm3 is a sort of cosmic prison, and life within the universe
in such a state is inescapably condemned to extinction as a result of proton decay and
the exponential falloff of the density of matter when the vacuum energy V(ϕ0 ) becomes
dominant. There is still no consensus on the probability of the spontaneous emergence of
complex life forms on Earth through just a single evolutionary chain. If, as some believe,
this probability is extremely low, and if some mechanism of indefinitely long reproduction
of life at V(ϕ0 ) < V∗ does actually exist (and this is not at all obvious a priori [336]), then
the existence of such a mechanism can drastically increase the fraction of the “habitable”
6
universes in a quantum state with V(ϕ0 ) < 10−5·10 g/cm3 as compared with the fraction
6
in a state with V(ϕ0 ) > 10−5·10 g/cm3 . The net result could then be [359] that any
observer like ourselves, who is capable of inquiring about the vacuum energy density,
would very likely find himself in a universe corresponding to a quantum state |αi i with

V(ϕ0 ) ≪ 10−29 g/cm3 .

Our treatment of this problem has merely provided a sketch of the course of future
research, illustrating the novel possibilities which have emerged in recent years in elemen-
tary particle theory and cosmology. If we are to develop a successful strategy for the
survival of mankind (if such a strategy exists), we will need to undertake a much deeper
study of the global structure of the inflationary universe and the conditions required for
the emergence and/or propagation of life therein. In any event, however, the possibility
that there is a the correlation between the value of the vacuum energy density and the
existence of a mechanism of eternal reproduction of life in the universe seems to us to be
noteworthy.
 Conclusion

Elementary particle theorists and cosmologists might be compared to two teams tunneling
toward each other through the enormous mountain of the unknown. The analogy, however,
is not entirely accurate. If the two teams of construction workers miss each other, they will
simply have built two tunnels instead of one. But in our case, if the particle theorists fail
to meet the cosmologists, we wind up without any complete theory at all. Furthermore,
even if they do meet, and manage to build an internally consistent theory of all processes
in the micro- and macro-worlds, that still does not mean that their theory is correct.
In the face of the now familiar (and inevitable) dearth of experimental data on particle
interactions at energies approaching 1019 GeV, and on the structure of the universe on
scales l ≫ 1028 cm, it becomes especially important to guess, if only in broad outline, the
true direction that this science will take, a direction that should remain valid even if many
specific details of the theory under construction should change. This explains the recent
emergence of such unusual terms as scenario and paradigm in the physicists’ lexicon.
The major developments in elementary particle physics over the past two decades can
be characterized by a few key words, such as gauge invariance, unified theories with
spontaneous symmetry breaking, supersymmetry, and strings. The term inflation has
become such a word in the cosmology of the 1980’s.
The inflationary universe scenario could only have been created
through the joint efforts of cosmologists and elementary particle theorists. The need
for and fruitfulness of such a collaboration is now obvious. It should be pointed out that
inflation is certainly not a magic word that will automatically solve all our problems and
open all doors. In some theories of elementary particles, it is difficult to implement the
inflationary universe scenario, whereas many other theories fail to lead to a good cosmol-
ogy, even with the help of inflation. The road to a consistent cosmological theory may
yet prove to be a very long one, and we may still find that many details of our present
scenarios will be cast off as inessential excess baggage. At the moment, however, it does
seem necessary to have something like inflation to obtain a consistent cosmology at peace
with particle physics.
Inflationary cosmology continues to develop rapidly. We are witnessing a gradual
change in our most general concepts about the evolution of the universe. Just a few years
ago, most authorities had virtually no doubt that the universe was born in a unique Big
 244

Bang approximately 10–15 billion years in the past. It seemed obvious that space-time
was four-dimensional from the very outset, and that it remains four-dimensional all over
the universe. It was believed that if the universe were closed, its total size could hardly
exceed that of its observable part, l ∼ 1028 cm, and that in no more than 1011 years such
a universe would collapse and disappear. If, on the other hand, the universe were open or
flat, then it would be infinite, and the general conviction was that it would then exhibit
properties everywhere that were almost the same as those in its observable part. Such a
universe would exist forever, but after its protons had decayed, as predicted by unified
theories of the weak, strong, and electromagnetic interactions, no baryon matter would
remain to support life. The only alternatives, then, were a “hot end” with the expected
collapse of the universe, and a “cold end” in the infinite void of space.
It now seems more likely that the universe as a whole exists eternally, endlessly spawn-
ing ever newer exponentially large regions in which the low-energy laws governing elemen-
tary particle interactions, and even the effective dimensionality of space-time itself, may
be different. We do not know whether life can evolve forever in each such region , but
we do know for certain that life will appear again and again in different regions of the
universe, taking on all possible forms. This change in our notions of the global structure
of the universe and our place within it is one of the most important consequences to come
out of inflationary cosmology.
We have finally come to a newfound appreciation of why it was necessary to write the
scenario even though the performance was already over — the show is still going on, and
most likely will continue forever. In different parts of the universe, different audiences will
observe it in its infinite variations. We cannot witness the whole play in all its grandeur,
but we can try to imagine its most essential features — and ultimately perhaps even grasp
its meaning.
 References

1. S. L. Glashow, Nucl. Phys. 22, 579 (1961);

S. Weinberg, Phys. Rev. Lett. 19, 1264 (1967);
A. Salam, in: Elementary Particle Theory, edited by N. Svartholm, Almquist and
Wiksell, Stockholm (1968), p. 367.
2. G. ’t Hooft, Nucl. Phys. B35, No. 1, 167–188 (1971);
B. W. Lee, Phys. Rev. D5, No. 4, 823–835 (1972);
B. W. Lee and J. Zinn–Justin, Phys. Rev. D5, No. 12, 3121–3160 (1972);
G. ’t Hooft and M. Veltman, Nucl. Phys. B50, No. 2, 318–335 (1972);
I. V. Tyutin and E. S. Fradkin, Sov. J. Nucl. Phys. 16, 464 (1972);
R. E. Kallosh and I. V. Tyutin, Sov. J. Nucl. Phys. 17, 98 (1973).
3. D. J. Gross and F. Wilczek, Phys. Rev. Lett. 30, No. 26, 1343–1346 (1973);
H. D. Politzer, Phys. Rev. Lett. 30, No. 26, 1346–1349 (1973).
4. H. Georgi and S. L. Glashow, Phys. Rev. Lett. 32, 438 (1974).
5. D. Z. Friedman, P. van Nieuwenhuizen, and S. Ferrara, Phys. Rev. D13, 3214 (1976).
6. Th. Kaluza, Stizungsber. Preuss. Akad. Wiss. Phys., Math., K1, 966 (1921);
O. Klein, Z. Phys. 37, 895 (1926);
E. Cremmer and J. Scherk, Nucl. Phys. B 108, 409;
E. Witten, Nucl. Phys. B 186, 412 (1981).
7. M. B. Green and J. H. Schwarz, Phys. Lett. 149B, 117 (1984);
Phys. Lett. 151B, 21 (1984);
D. J. Gross, J. A. Harvey, E. Martinec, and R. Rohm, Phys. Rev. Lett. 54, 502
(1985);
E. Witten, Phys. Lett. 149B, 351 (1984).
8. A. A. Slavnov and L. D. Faddeev, Gauge Fields: Introduction to Quantum Theory,
Benjamin–Cummings, London (1980).
9. J. C. Taylor, Gauge Theories of Weak Interactions, Cambridge University Press,
Cambridge (1976).
10. L. B. Okun, Leptons and Quarks, North–Holland, Amsterdam (1982).
11. A. M. Polyakov, Gauge Fields and Strings, Harwood, London (1987).
12. P. Langacker, Phys. Rep. C 72, 185 (1981).
REFERENCES 246

13. V. I. Ogievetsky and L. Mezincescu, Sov. Phys. Usp. 18, 960 (1975);
J. Bagger and J. Wess, Supersymmetry and Supergravity, Princeton Univ. Press,
Princeton (1983);
P. West, Introduction to Supersymmetry and Supergravity, World Scientific, Singa-
pore (1986).
14. P. van Nieuwenhuizen, Phys. Rep. C 68, 192 (1981).
15. H. P. Nilles, Phys. Rep. C 110, 3 (1984).
16. M. J. Duff, B. E. W. Nilsson, and C. N. Pope, Phys. Rep. C 130, 1 (1986).
17. M. B. Green, J. H. Schwarz, and E. Witten, Superstring Theory, Cambridge Uni-
versity Press, Cambridge (1987).
18. D. A. Kirzhnits, JETP Lett. 15, 529 (1972).
19. D. A. Kirzhnits and A. D. Linde, Phys. Lett. 42B, 471 (1972).
20. S. Weinberg, Phys. Rev. D9, 3320 (1974);
L. Dolan and R. Jackiw, Phys. Rev. D9, 3357 (1974).
21. D. A. Kirzhnits and A. D. Linde, Sov. Phys. JETP 40, 628 (1974).
22. D. A. Kirzhnits and A. D. Linde, Lebedev Phys. Inst. Preprint No. 101 (1974).
23. D. A. Kirzhnits and A. D. Linde, Ann. Phys. 101, 195 (1976).
24. A. D. Linde, Rep. Prog. Phys. 42, 389 (1979).
25. T. D. Lee and G. C. Wick, Phys. Rev. D9, 2291 (1974).
26. B. J. Harrington and A. Yildis, Phys. Rev. Lett. 33, 324 (1974).
27. A. D. Linde, Phys. Rev. D14, 3345 (1976);
I. V. Krive, A. D. Linde, and E. M. Chudnovsky, Sov. Phys. JETP 44, 435 (1976).
28. A. D. Linde, Phys. Lett. 86B, 39 (1979).
29. I. V. Krive, Sov. Phys. JETP 56, 477 (1982).
30. A. Salam and J. Strathdee, Nature 252, 569 (1974);
A. Salam and J. Strathdee, Nucl. Phys. B90, 203 (1975).
31. A. D. Linde, Phys. Lett. 62B, 435 (1976).
32. I. V. Krive, V. M. Pyzh, and E. M. Chudnovsky, Sov. J. Nucl. Phys. 23, 358 (1976).
33. V. V. Skalozub, Sov. J. Nucl. Phys. 45, 1058 (1987).
34. Ya. B. Zeldovich and I. D. Novikov, Structure and Evolution of the Universe [in
Russian], Nauka, Moscow (1975) [transl. as Relativistic Astrophysics, Vol. II, Univ.
of Chicago Press, Chicago (1983)].
35. S. Weinberg, Gravitation and Cosmology: Principles and Applications of the General
Theory of Relativity, J. Wiley and Sons, New York (1972).
36. A. D. Sakharov, JETP Lett. 5, 24 (1967).
37. V. A. Kuzmin, JETP Lett. 12, 335 (1970).
38. A. Yu. Ignatiev, N. V. Krasnikov, V. A. Kuzmin, and A. N. Tavkhelidze, Phys. Lett.
76B, 436 (1978);
M. Yoshimura, Phys. Rev. Lett. 41, 281 (1978);
S. Weinberg, Phys. Lett. 42, 850 (1979);
A. D. Dolgov, JETP Lett. 29, 228 (1979);
W. Kolb and S. Wolfram, Nucl. Phys. B 172, 224 (1980).
REFERENCES 247

39. Ya. B. Zeldovich, in: Magic without Magic: John Archibald Wheeler, a Collection
of Essays in Honor of His 60th Birthday, edited by J. R. Klauder, W. H. Freeman,
San Francisco (1972).
40. Ya. B. Zeldovich and M. Yu. Khlopov, Phys. Lett. 79B, 239 (1978);
J. P. Preskill, Phys. Rev. Lett. 43, 1365 (1979).
41. Ya. B. Zeldovich, I. Yu. Kobzarev, and L. B. Okun, Phys. Lett. 50B, 340 (1974).
42. S. Parke and S. Y. Pi, Phys. Lett. 107B, 54 (1981);
G. Lazarides, Q. Shafi, and T. F. Walsh, Nucl. Phys. B 195, 157 (1982).
43. P. Sikivie, Phys. Rev. Lett. 48, 1156 (1982).
44. J. Ellis, A. D. Linde, and D. V. Nanopoulos, Phys. Lett. 128B, 295 (1983).
45. M. Yu. Khlopov and A. D. Linde, Phys. Lett. 138B, 265 (1982).
46. J. Polonyi, Budapest Preprint KFKI-93 (1977).
47. G. D. Coughlan, W. Fischler, E. W. Kolb, S. Raby, and G. G. Ross, Phys. Lett.
131B, 59 (1983).
48. A. S. Goncharov, A. D. Linde, and M. I. Vysotsky, Phys. Lett. 147B, 279 (1984).
49. J. P. Preskill, M. B. Wise, and F. Wilczek, Phys. Lett. 120B, 127 (1983);
L. F. Abbott and P. Sikivie, Phys. Lett. 120B, 133 (1983);
M. Dine and W. Fischler, Phys. Lett. 120B, 133 (1983).
50. K. Choi and J. E. Kim, Phys. Lett. 154B, 393 (1985).
51. E. B. Gliner, Sov. Phys. JETP 22, 378 (1965);
E. B. Gliner, Docl.Akad.Nauk SSSR 192, 771 (1970);
E. B. Gliner and I. G. Dymnikova, Sov. Astron. Lett. 1, 93 (1975).
52. A. A. Starobinsky, JETP Lett. 30, 682 (1979);
A. A. Starobinsky, Phys. Lett. B 91, 99 (1980).
53. A. H. Guth, Phys. Rev. D23, 347 (1981).
54. A. D. Linde, Phys. Lett. 108B, 389 (1982).
55. A. Albrecht and P. J. Steinhardt, Phys. Rev. Lett. 48, 1220 (1982).
56. A. D. Linde, JETP Lett. 38, 149 (1983);
A. D. Linde, Phys. Lett. 129B, 177 (1983).
57. A. D. Linde, Mod. Phys. Lett. 1A, 81 (1986);
A. D. Linde, Phys. Lett. 175B, 395 (1986);
A. D. Linde, Physica Scripta T15, 169 (1987).
58. N. N. Bogolyubov and D. V. Shirkov, Introduction to the Theory of Quantized
Fields, Wiley, New York (1980).
59. P. W. Higgs, Phys. Rev. Lett. 13, 508 (1964);
T. W. B. Kibble, Phys. Rev. 155, 1554 (1967);
G. S. Guralnik, C. R. Hagen, and T. W. B. Kibble, Phys. Rev. Lett. 13, 585 (1964);
F. Englert and R. Brout, Phys. Rev. Lett. 13, 321 (1964).
60. V. L. Ginzburg and L. D. Landau, Sov. Phys. JETP 20, 1064 (1950).
61. L. D. Landau and E. M. Lifshitz, Statistical Physics, Pergamon, London (1968).
62. A. D. Linde, Phys. Lett. 100B, 37 (1981);
A. D. Linde, Nucl. Phys. B 216, 421 (1983).
63. A. Friedmann, Z. Phys. 10, 377 (1922).
REFERENCES 248

64. H. P. Robertson, Rev. Mod. Phys. 5, 62 (1933);

A. G. Walker, J. London Math. Soc. 19, 219 (1944).
65. L. D. Landau and E. M. Lifshitz, The Classical Theory of Fields, Pergamon, Oxford
(1975).
66. G. Gamow, Phys. Rev. 74, 505 (1948).
67. A. G. Doroshkevich and I. D. Novikov, Dokl. Akad. Nauk SSSR 154, 809 (1964).
68. V. A. Belinsky, E. M. Lifshitz, and I. M. Khalatnikov, Sov. Phys. Usp. 13, 457
(1971).
69. R. Penrose, “Structure of Space-Time”, in: Battelle Rencontres, 1967 Lectures in
Mathematics and Physics, edited by C. M. DeWitt and J. A. Wheeler, Benjamin,
New York (1968).
70. S. W. Hawking and G. F. Ellis, The Large Scale Structure of Space-Time, Cambridge
University Press, Cambridge (1973).
71. J. A. Wheeler, in: Relativity, Groups and Topology, edited by B. S. DeWitt and
C. M. DeWitt, Gordon and Breach, New York (1964);
S. W. Hawking, Nucl. Phys. B 144, 349 (1978).
72. C. W. Misner, Phys. Rev. Lett. 28, 1669 (1972).
73. C. B. Collins and S. W. Hawking, Astrophys. J. 180, 317 (1973).
74. A. A. Grib, S. G. Mamaev, and V. M. Mostepanenko, Quantum Effects in Strong
External Fields [in Russian], Atomizdat, Moscow (1980);
N. Birrell and P. C. Davies, Quantum Fields in Curved Space, Cambridge University
Press, Cambridge (1984).
75. E. M. Lifshitz, Sov. Phys. JETP 16, 587 (1946).
76. Ya. B. Zeldovich, Mon. Not. R. Astron. Soc. 160, 1 (1970).
77. R. H. Dicke, Nature 192, 440 (1961);
A. A. Zelmanov, in: Infinity and the Universe [in Russian], Mysl’, Moscow (1969),
p. 274;
B. Carter, in: Confrontation of Cosmological Theories with Observational Data,
edited by M. S. Longair, D. Reidel, Dordrecht (1974);
B. J. Carr and M. J. Rees, Nature 278, 605 (1979);
I. L. Rozental, Big Bang Big Bounce: How Particles and Fields Drive Cosmic Evo-
lution, Springer-Verlag, Berlin (1988).
78. A. D. Linde, in: 300 Years of Gravitation, edited by S. W. Hawking and W. Israel,
Cambridge University Press, Cambridge (1987), p. 604.
79. A. D. Linde, Phys. Today 40, 61 (1987).
80. S. Weinberg, Phys. Rev. Lett. 36, 294 (1976).
81. A. Vilenkin, Phys. Rep. 121, 263 (1985).
82. G. ’t Hooft, Nucl. Phys. B79, 279 (1974).
83. A. M. Polyakov, JETP Lett. 20, 430 (1974).
84. T. W. B. Kibble, J. Phys. 9A, 1387 (1976).
85. Yu. A. Golfand and E. P. Likhtman, JETP Lett. 13, 323 (1971);
J. Gervais and B. Sakita, Nucl. Phys. B34, 632 (1971);
REFERENCES 249

D. V. Volkov and V. P. Akulov, JETP Lett. 16, 438 (1972);

J. Wess and B. Zumino, Nucl. Phys. B70, 39 (1974).
86. J. Ellis and D. V. Nanopoulos, Phys. Lett. 116B, 133 (1982).
87. A. B. Lahanas and D. V. Nanopoulos, Phys. Rep. 145, 3 (1987).
88. A. D. Linde, JETP Lett. 19, 183 (1974);
M. Veltman, Rockefeller University Preprint (1974);
M. Veltman, Phys. Rev. Lett. 34, 77 (1975);
J. Dreitlein, Phys. Rev. Lett. 33, 1243 (1975).
89. Ya. B. Zeldovich, Sov. Phys. Usp. 11, 381 (1968).
90. A. Einstein, Über den gegenwärtigen Stand der Feld-Theorie. Festschrift
Prof. Dr. A. Stodola zum 70. (Geburstag, Füssle Verlag, Zurich u. Leipzig, 1929)
p. 126.
91. E. S. Fradkin, in: Proceedings of the Quark-80 Seminar, Moscow (1980);
S. Dimopoulos and H. Georgi, Nucl. Phys. B 193, 150 (1981);
N. Sakai, Z. Phys. C 11, 153 (1981).
92. N. V. Dragon, Phys. Lett. 113B, 288 (1982);
P. H. Frampton and T. W. Kephart, Phys. Rev. Lett. 48, 1237 (1982);
F. Buccella, J. P. Deredinger, S. Ferrara, and C. A. Savoy, Phys. Lett. 115B, 375
(1982).
93. D. V. Nanopoulos and K. Tamvakis, Phys. Lett. B 110, 449 (1982);
M. Srednicki, Nucl. Phys. B 202, 327 (1982).
94. P. Freund, Phys. Lett. 151B, 387 (1985);
A. Casher, F. Englert, H. Nicolai, and A. Taormina, Phys. Lett. 162B, 121 (1985).
95. M. J. Duff, B. E. W. Nilsson, and C. N. Pope, Phys. Lett. 163B, 343 (1985).
96. R. E. Kallosh, Phys. Lett. 176B, 50 (1986);
R. E. Kallosh, Physica Scripta T15, 118 (1987).
97. J. Affleck and M. Dine, Nucl. Phys. B249, 361 (1985).
98. A. D. Linde, Phys. Lett. 160B, 243 (1985).
99. S. Dimipoulos and L. J. Hall, Phys. Lett. 196B, 135 (1987).
100. W. de Sitter, Proc. Kon. Ned. Akad. Wet. 19, 1217 (1917);
W. de Sitter, Proc. Kon. Ned. Akad. Wet. 20, 229 (1917).
101. A. D. Sakharov, Sov. Phys. JETP 22, 241 (1965).
102. B. L. Altshuler, in: Abstracts from the Third Soviet Conference on Gravitation [in
Russian], Erevan (1972), p. 6.
103. L. E. Gurevich, Astrophys. Space Sci. 38, 67 (1975).
104. A. D. Linde, Phys. Lett. 99B, 391 (1981).
105. A. D. Dolgov and Ya. B. Zeldovich, Rev. Mod. Phys. 53, 1 (1981).
106. J. S. Dowker and R. Critchley, Phys. Rev. D13, 3224 (1976).
107. V. F. Mukhanov and G. V. Chibisov, JETP Lett. 33, 523 (1981);
V. F. Mukhanov and G. V. Chibisov, Sov. Phys. JETP 56, 258 (1982).
108. J. D. Barrow and A. Ottewill, J. Phys. A16, 2757 (1983).
109. A. A. Starobinsky, Sov. Astron. Lett. 9, 302 (1983).
110. L. A. Kofman, A. D. Linde, and A. A. Starobinsky, Phys. Lett. 157B, 36 (1985).
REFERENCES 250

111. V. G. Lapchinsky, V. A. Rubakov, and A. V. Veryaskin, Inst. Nucl. Res. Preprint

No. P-0195 (1982).
112. S. W. Hawking, I. G. Moss, and J. M. Stewart, Phys. Rev. D26, 2681 (1982).
113. A. H. Guth and E. J. Weinberg, Nucl. Phys. B212, 321 (1983).
114. S. W. Hawking, Phys. Lett. 115B, 295 (1982);
A. A. Starobinsky, Phys. Lett. 117B, 175 (1982);
A. H. Guth and S. Y. Pi, Phys. Rev. Lett. 49, 1110 (1982);
J. M. Bardeen, P. J. Steinhardt, and M. S. Turner, Phys. Rev. D28, 679 (1983).
115. A. D. Linde, Phys. Lett. 132B, 317 (1983).
116. A. D. Linde, Rep. Prog. Phys. 47, 925 (1984).
117. V. A. Rubakov, M. V. Sazhin, and A. V. Veryaskin, Phys. Lett. 115B, 189 (1982).
118. A. D. Linde, Phys. Lett. 162B, 281 (1985);
A. D. Linde, Suppl. Prog. Theor. Phys. 85, 279 (1985).
119. I. D. Novikov and V. P. Frolov, The Physics of Black Holes [in Russian], Nauka,
Moscow (1986).
120. G. W. Gibbons and S. W. Hawking, Phys. Rev. D15, 2738 (1977).
121. S. W. Hawking and I. G. Moss, Phys. Lett. 110B, 35 (1982).
122. W. Boucher and G. W. Gibbons, in: The Very Early Universe, edited by G. W. Gib-
bons, S. W. Hawking, and S. Siklos, Cambridge University Press, Cambridge (1983);
A. A. Starobinsky, JETP Lett. 37, 66 (1983);
R. Wald, Phys. Rev. D28, 2118 (1983);
E. Martinez-Gonzalez and B. J. T. Jones, Phys. Lett. 167B, 37 (1986);
I. G. Moss and V. Sahni, Phys. Lett. B178, 159 (1986);
M. S. Turner and L. Widrow, Phys. Rev. Lett. 57, 2237 (1986);
L. Jensen and J. Stein-Schabes, Phys. Rev. D34, 931 (1986).
123. A. D. Dolgov and A. D. Linde, Phys. Lett. 116B, 329 (1982).
124. L. F. Abbott, E. Farhi, and M. B. Wise, Phys. Lett. 117B, 29 (1982).
125. L. A. Kofman and A. D. Linde, Nucl. Phys. B282, 555 (1987).
126. A. Vilenkin and L. H. Ford, Phys. Rev. D26, 1231 (1982).
127. A. D. Linde, Phys. Lett. 116B, 335 (1982).
128. A. A. Starobinsky, Phys. Lett. 117B, 175 (1982).
129. V. A. Kuzmin, V. A. Rubakov, and M. E. Shaposhnikov, Phys. Lett. 115B, 36
(1985).
130. M. E. Shaposhnikov, JETP Lett. 44, 465 (1986);
Nucl. Phys. B299, 797 (1988);
L. McLerran, Phys. Rev. Lett. 62, 1075 (1989).
131. A. Dannenberg and L. J. Hall, Phys. Lett. 198B, 411 (1987).
132. A. S. Goncharov and A. D. Linde, Sov. Phys. JETP 65, 635 (1987).
133. A. S. Goncharov, A. D. Linde, and V. F. Mukhanov, Int. J. Mod. Phys. 2A, 561
(1987).
134. A. A. Starobinsky, in: Funadamental Interactions [in Russian], MGPI, Moscow
(1984), p. 55.
REFERENCES 251

135. A. A. Starobinsky, in: Current Trends in Field Theory, Quantum Gravity, and
Strings, Lecture Notes in Physics, edited by H. J. de Vega and N. Sanches, Springer-
Verlag, Heidelberg (1986).
136. L. P. Grishchuk and Ya. B. Zeldovich, Sov. Astron. 22, 12 (1978).
137. S. Coleman and E. J. Weinberg, Phys. Rev. D6, 1888 (1973).
138. R. Jackiw, Phys. Rev. D9, 1686 (1973).
139. A. D. Linde, JETP Lett. 23, 73 (1976).
140. S. Weinberg, Phys. Rev. Lett. 36, 294 (1976).
141. A. D. Linde, Phys. Lett. 70B, 306 (1977).
142. A. D. Linde, Phys. Lett. 92B, 119 (1980).
143. A. H. Guth and E. J. Weinberg, Phys. Rev. Lett. 45, 1131 (1980).
144. E. Witten, Nucl. Phys. B177, 477 (1981).
145. I. V. Krive and A. D. Linde, Nucl. Phys. B117, 265 (1976).
146. A. D. Linde, Trieste Preprint No. IC/76/26 (1976).
147. N. V. Krasnikov, Sov. J. Nucl. Phys. 28, 279 (1978).
148. P. Q. Hung, Phys. Rev. Lett. 42, 873 (1979).
149. H. D. Politzer and S. Wolfram, Phys. Lett. 82B, 242 (1979).
150. A. A. Anselm, JETP Lett. 29, 645 (1979).
151. N. Cabibbo, L. Maiani, A. Parisi, and R. Petronzio, Nucl. Phys. B158, 295 (1979).
152. B. L. Voronov and I. V. Tyutin, Sov. J. Nucl. Phys. 23, 699 (1976).
153. S. Coleman, R. Jackiw, and H. D. Politzer, Phys. Rev. D10, 2491 (1974).
154. L. F. Abbott, J. S. Kang, and H. J. Schnitzer, Phys. Rev. D13, 2212 (1976).
155. A. D. Linde, Nucl. Phys. B125, 369 (1977).
156. L. D. Landau and I. Ya. Pomeranchuk, Dokl. Akad. Nauk SSSR 102, 489 (1955).
157. E. S. Fradkin, Sov. Phys. JETP 28, 750 (1955).
158. A. B. Migdal, Fermions and Bosons in Strong Fields [in Russian], Nauka, Moscow
(1978).
159. D. A. Kirzhnits and A. D. Linde, Phys. Lett. 73B, 323 (1978).
160. J. Fröhlich, Nucl. Phys. B200, 281 (1982).
161. C. B. Lang, Nucl. Phys. B265, 630 (1986).
162. D. A. Kirzhnits, in: Quantum Field Theory and Quantum Statistics, edited by
I. A. Batalin, C. J. Isham, and G. A. Vilkovisky, Adam Hilger Press, Bristol (1987),
Vol. 1, p. 349.
163. W. A. Bardeen and M. Moshe, Phys. Rev. D28, 1372 (1982).
164. K. Enquist and J. Maalampi, Phys. Lett. 180B, 14 (1986).
165. L. Smolin, Phys. Lett. 93B, 95 (1980).
166. E. S. Fradkin, Proc. Lebedev Phys. Inst. 29, 7 (1965) [English transl.: Consultants
Bureau, New York (1967)].
167. V. A. Kuzmin, M. E. Shaposhnikov, and I. I. Tkachev, Z. Phys. Ser. C 12, 83 (1982).
168. M. B. Kislinger and P. D. Morley, Phys. Rev. D13, 2765 (1976).
169. E. V. Shuryak, Sov. Phys. JETP 47, 212 (1978).
170. A. M. Polyakov, Phys. Lett. 72B, 477 (1978).
171. A. D. Linde, Phys. Lett. 96B, 289 (1980).
REFERENCES 252

172. D. J. Gross, R. Pisarski, and L. Yaffe, Rev. Mod. Phys. 53, 43 (1981).
173. A. D. Linde, Phys. Lett. 96B, 293 (1980).
174. M. A. Matveev, V. A. Rubakov, A. N. Tavkhelidze, and V. F. Tokarev, Nucl. Phys.
B282, 700 (1987).
175. D. I. Deryagin, D. Ya. Grigoriev, and V. A. Rubakov, Phys. Lett. 178B, 385 (1986).
176. O. K. Kalashnikov and H. Perez–Rojas, Nucl. Phys. B293, 241 (1987).
177. E. J. Ferrer, V.de la Incera, and A. E. Shabad, Phys. Lett. 185B, 407 (1987);
E. J. Ferrer and V. de la Incera, Phys. Lett. 205B, 381 (1988).
178. M. E. Shaposhnikov, Nucl. Phys. B287, 767 (1987).
179. M. B. Voloshin, I. B. Kobzarev, and L. B. Okun, Sov. J. Nucl. Phys. 20, 644 (1974).
180. S. Coleman, Phys. Rev. D15, 2929 (1977).
181. C. Callan and S. Coleman, Phys. Rev. D16, 1762 (1977).
182. S. Fubini, Nuovo Cimento 34A, 521 (1976).
183. G. ’t Hooft, Phys. Rev. D14, 3432 (1976);
R. Rajaraman, Solitons and Instantons in Quantum Field Theory, North-Holland,
Amsterdam (1982).
184. N. V. Krasnikov, Phys. Lett. 72B, 455 (1978).
185. J. Affleck, Nucl. Phys. B191, 429 (1981).
186. A. S. Goncharov and A. D. Linde, Sov. J. Part. Nucl. 17, 369 (1986).
187. A. D. Linde, “The kinetics of phase transitions in grand unified theories” [in Russian],
FIAN Preprint No. 266 (1981).
188. R. Flores and M. Sher, Phys. Rev. D27, 1679 (1983).
189. M. Sher and H. W. Zaglauer, Phys. Lett 206B, 527 (1988).
190. A. A. Abrikosov, Sov. Phys. JETP 32, 1442 (1957);
H. B. Nielsen and P. Olesen, Nucl. Phys. B61, 45 (1973).
191. Ya. B. Zeldovich, Mon. Not. R. Astron. Soc. 192, 663 (1980).
192. A. Vilenkin, Phys. Rev. Lett. 46, 1169 (1981).
193. N. Turok and R. Brandenberger, Phys. Rev. D33, 2175 (1986).
194. E. N. Parker, Astrophys. J. 160, 383 (1970).
195. E. W. Kolb, S. A. Colgate, and J. A. Harvey, Phys. Rev. Lett. 49, 1373 (1982);
S. Dimopoulos, J. Preskill, and F. Wilczek, Phys. Lett. 119B, 320 (1982);
K. Freese, M. S. Turner, and D. N. Schramm, Phys. Rev. Lett. 51, 1625 (1983).
196. V. A. Rubakov, Nucl. Phys. B203, 311 (1982);
C. G. Callan, Phys. Rev. D26, 2058 (1982).
197. Y. Nambu, Phys. Rev. D10, 4262 (1974).
198. A. Billoire, G. Lazarides, and Q. Shafi, Phys. Lett. 103B, 450 (1981).
199. T. A. DeGrand and D. Toussaint, Phys. Rev. D25, 526 (1982).
200. V. N. Namiot and A. D. Linde, unpublished.
201. G. W. Gibbons and S. W. Hawking, Phys. Rev. D15, 2752 (1977).
202. T. S. Bunch and P. C. W. Davies, Proc. R. Soc. London, Ser. A A360, 117 (1978).
203. A. Vilenkin, Nucl. Phys. B226, 527 (1983).
204. A. Vilenkin, Phys. Rev. D27, 2848 (1983).
205. Yu. L. Klimontovich, Statistical Physics [in Russian], Nauka, Moscow (1982).
REFERENCES 253

206. S.-J. Rey, Nucl. Phys. B284, 706 (1987).

207. S. Coleman and F. De Luccia, Phys. Rev. D21, 3305 (1980).
208. N. Deruelle, Mod. Phys. Lett. 4A, 1297 (1989).
209. S. W. Hawking and I. G. Moss, Nucl. Phys. B224, 180 (1983).
210. H. A. Kramers, Physica 7, 240 (1940).
211. A. D. Linde, Phys. Lett. 131B, 330 (1983).
212. W. Israel, Nuovo Cimento 44B, 1 (1966).
213. V. A. Berezin, V. A. Kuzmin, and I. I. Tkachev, Phys. Lett. 120B, 91 (1983);
V. A. Berezin, V. A. Kuzmin, and I. I. Tkachev, Phys. Rev. D36, 2919 (1987);
A. Aurilia, G. Denardo, F. Legovini, and E. Spalucci, Nucl. Phys. B252, 523 (1985);
P. Laguna-Gastillo, and R. A. Matzner, Phys. Rev. D34, 2913 (1986);
S. K. Blau, E. I. Guendelman, and A. H. Guth, Phys. Rev. D35, 1747 (1987);
A. Aurilia, R. S. Kissack, R. Mann, and E. Spalucci, Phys. Rev. D35, 2961 (1987).
214. E. R. Harrison, Phys. Rev. D1, 2726 (1973).
215. L. P. Grishchuk, Sov. Phys. JETP 40, 409 (1974).
216. V. N. Lukash, Sov. Phys. JETP 52, 807 (1980);
D. A. Kompaneets, V. N. Lukash, and I. D. Novikov, Sov. Astron. 26, 259 (1982);
V. N. Lukash and I. D. Novikov, in: The Very Early Universe, edited by G. W. Gib-
bons, S. W. Hawking, and S. Siklos, Cambridge University Press, Cambridge (1983),
p. 311.
217. V. F. Mukhanov and G. V. Chibisov, Mon. Not. Astron. Soc. 200, 535 (1982).
218. V. F. Mukhanov, JETP Lett. 41, 493 (1985).
219. R. H. Brandenberger, Rev. Mod. Phys. 57, 1 (1985);
R. H. Brandenberger, Int. J. Mod. Phys. 2A, 77 (1987).
220. J. M. Bardeen, Phys. Rev. D22, 1882 (1980);
G. V. Chibisov and V. F. Mukhanov, Lebedev Phys. Inst. Preprint No. 154 (1983).
221. L. A. Kofman, V. F. Mukhanov, and D. Yu. Pogosyan, Sov. Phys. JETP 66, 441
(1987).
222. V. F. Mukhanov, L. A. Kofman, and D. Yu. Pogosyan, Phys. Lett. 157B, 427 (1987).
223. P. J. E. Peebles, Astrophys. J. Lett. 263, L1 (1982).
224. S. F. Shandarin, A. G. Doroshkevich, and Ya. B. Zeldovich, Sov. Phys. Usp. 26, 46
(1983).
225. A. A. Starobinsky, Sov. Astron. Lett. 9, 302 (1983).
226. V. N. Lukash, P. D. Naselskij, and I. D. Novikov, in: Quantum Gravity, edited by
M. A. Markov, V. A. Berezin, and V. P. Frolov, World Scientific, Singapore (1984),
p. 675;
V. N. Lukash and I. D. Novikov, Nature 316, 46 (1985).
227. L. A. Kofman and A. A. Starobinsky, Sov. Astron. Lett. 11, 643 (1985);
L. A. Kofman, D. Yu. Pogosyan, and A. A. Starobinsky, Sov. Astron. Lett. 12, 419
(1985).
228. A. B. Berlin, E. V. Bulaenko, V. V. Vitkovsky, V. K. Kononov, Yu. N. Parijskij, and
Z. E. Petrov, Proc. IAU Symposium 104, edited by G. O. Abell and G. Chincarini,
D. Reidel, Dordrecht (1983);
REFERENCES 254

F. Melchiorri, B. Melchiorri, C. Ceccarelli, and L. Pietranera, Astrophys. J. Lett.

250, L1 (1981);
I. A. Strukov and D. P. Skulachev, Sov. Astron. Lett. 10, 1 (1984);
J. M. Uson and D. T. Wilkinson, Nature 312, 427 (1984);
R. D. Davies et al., Nature 326, 462 (1987);
D. J. Fixen, E. S. Cheng, and D. T. Wilkinson, Phys. Rev. Lett. 50, 620 (1983).
229. D. H. Lyth, Phys. Lett. 147B, 403 (1984);
S. W. Hawking, Phys. Lett. 150B, 339 (1985).
230. C. W. Kim, and P. Murphy, Phys. Lett. 167B, 43 (1986).
231. S. W. Hawking, in: 300 Years of Gravitation, edited by S. W. Hawking and W. Israel,
Cambridge University Press, Cambridge (1987), p. 631.
232. H. Georgi and S. L. Glashow, Phys. Rev. Lett. 28, 1494 (1972).
233. R. D. Peccei and H. Quinn, Phys. Rev. Lett. 38, 1440 (1977);
R. D. Peccei and H. Quinn, Phys. Rev. D16, 1791 (1977).
234. S. Weinberg, Phys. Rev. Lett. 40, 223 (1978);
F. Wilczek, Phys. Rev. Lett. 40, 279 (1978).
235. J. R. Primack, in: Proc. of the International School of Physics “Enrico Fermi” (1984);
M. J. Rees, in: 300 Years of Gravitation, edited by S. W. Hawking and W. Israel,
Cambridge University Press, Cambridge (1987).
236. A. G. Doroshkevich, Sov. Astron. Lett. 14, 125 (1988).
237. Q. Shafi and C. Wetterich, Phys. Lett. 152B, 51 (1985);
Q. Shafi and C. Wetterich, Nucl. Phys. B289, 787 (1987).
238. J. Silk and M. S. Turner, Phys. Rev. D35, 419 (1987).
239. A. D. Linde, JETP Lett. 40, 1333 (1984);
A. D. Linde, Phys. Lett. 158B, 375 (1985).
240. D. Seckel and M. S. Turner, Phys. Rev. D32, 3178 (1985).
241. L. A. Kofman, Phys. Lett. 174B, 400 (1986).
242. L. A. Kofman and D. Yu. Pogosyan, Phys. Lett. 214B, 508 (1988).
243. L. A. Kofman, A. D. Linde, and J. Einasto, Nature 326, 48 (1987).
244. J. Goldstone, Nuovo Cimento 19, 154 (1961);
J. Goldstone, A. Salam, and S. Weinberg, Phys. Rev. 127, 965 (1962).
245. T. J. Allen, B. Grinstein, and M. B. Wise, Phys. Lett. 197B, 66 (1987).
246. Q. Shafi and A. Vilenkin, Phys. Rev. D29, 1870 (1984).
247. E. T. Vishniac, K. A. Olive, and D. Seckel, Nucl. Phys. B289, 717 (1987).
248. A. D. Dolgov and N. S. Kardashov, Inst. of Space Research Preprint (1987);
A. D. Dolgov, A. F. Illarionov, N. S. Kardashov, and I. D. Novikov, Sov. Phys. JETP
67, 13 (1988).
249. V. de Lapparent, M. Geller, and J. Huchra, Astrophys. J. 302, L1 (1987).
250. J. P. Ostriker and L. Cowie, Astrophys. J. Lett. 243, L127 (1981).
251. I. I. Tkachev, Phys. Lett. 191B, 41 (1987).
252. M. S. Turner, Phys. Rev. D28, 1243 (1983).
253. R. J. Scherrer and M. S. Turner, Phys. Rev. D31, 681 (1985).
254. L. A. Kofman and A. D. Linde, to be published.
REFERENCES 255

255. Q. Shafi and C. Wetterich, Nucl. Phys. B297, 697 (1988).

256. A. Ringwald, Z. Phys. Ser. C 34, 481 (1987);
A. Ringwald, Heidelberg Preprint HD-THEP-85-18 (1985).
257. E. W. Kolb and M. S. Turner, Ann. Rev. Nucl. Part. Sci. 33, 645 (1983).
258. M. S. Turner, in: Architecture of Fundamental Interaction at Short Distances, edited
by P. Ramond and R. Stora, Elsevier Science Publishers, Copenhagen (1987).
259. V. A. Kuzmin, M. E. Shaposhnikov, and I. I. Tkachev, Nucl. Phys. B196, 29 (1982).
260. B. A. Campbell, J. Ellis, D. V. Nanopoulos, and K. A. Olive, Mod. Phys. Lett. A1,
389 (1986);
J. Ellis, D. V. Nanopoulos, and K. A. Olive, Phys. Lett. B184, 37 (1987);
J. Ellis, K. Enquist, D. V. Nanopoulos, and K. A. Olive, Phys. Lett. B191, 343
(1987).
261. M. Fukugita and T. Yanagida, Phys. Lett. 174B, 45 (1986).
262. K. Yamamoto, Phys. Lett. 168B, 341 (1986).
263. R. N. Mohaparta and J. W. F. Valle, Phys. Lett. 186B, 303 (1987).
264. G. M. Shore, Ann. Phys. 128, 376 (1980).
265. A. D. Linde, Phys. Lett. 114B, 431 (1982).
266. P. J. Steinhardt, in: The Very Early Universe, edited by G. W. Gibbons, S. W. Hawk-
ing, and S. Siklos, Cambridge University Press, Cambridge (1983).
267. A. D. Linde, “Nonsingular regenerating inflationary universe”, Cambridge University
Preprint (1982).
268. P. J. Steinhardt and M. S. Turner, Phys. Rev. D29, 2162 (1984).
269. A. Albrecht, S. Dimopoulos, W. Fischer, E. W. Kolb, S. Raby, and P. J. Steinhardt,
Nucl. Phys. B229, 528 (1983).
270. J. Ellis, D. V. Nanopoulos, K. A. Olive, and K. Tamvakis, Nucl. Phys. B221, 421
(1983);
D. V. Nanopoulos, K. A. Olive, M. Srednicki, and K. Tamvakis, Phys. Lett. 123B,
41 (1983).
271. B. Ovrut and P. J. Steinhardt, Phys. Rev. Lett. 53, 732 (1984);
B. Ovrut and P. J. Steinhardt, Phys. Lett. B147, 263 (1984).
272. E. Cremmer, S. Ferrara, L. Girardello, and A. Van Proeyen, Nucl. Phys. B212, 413
(1983).
273. A. S. Goncharov and A. D. Linde, Sov. Phys. JETP 59, 930 (1984);
A. S. Goncharov and A. D. Linde, Phys. Lett. 139B, 27 (1984).
274. A. S. Goncharov and A. D. Linde, Class. Quant. Grav. 1, L75 (1984).
275. Q. Shafi and A. Vilenkin, Phys. Rev. Lett. 52, 691 (1984).
276. S. Y. Pi, Phys. Rev. Lett. 52, 1725 (1984).
277. A. S. Goncharov, “Phase transitions in gauge theories and cosmology”, Candidate’s
Dissertation, Moscow (1984).
278. L. A. Khalfin, Sov. Phys. JETP 64, 673 (1986).
279. V. A. Belinsky, L. P. Grishchuk, Ya. B. Zeldovich, and I. M. Khalatnikov, Phys.
Lett. 155B, 232 (1985).
REFERENCES 256

280. V. A. Belinsky, and I. M. Khalatnikov, Sov. Phys. JETP 93, 784 (1987);
V. A. Belinsky, H. Ishihara, I. M. Khalatnikov, and H. Sato, Progr. Theor. Phys.
79, 676 (1988).
281. A. D. Linde, Phys. Lett. 202B, 194 (1988).
282. R. Holman, P. Ramond, and G. G. Ross, Phys. Lett. 137B, 343 (1984).
283. J. Ellis, A. B. Lahanas, D. V. Nanopoulos, and K. Tamvakis, Phys. Lett. 134B, 429
(1984);
J. Ellis, C. Kounnas, and D. V. Nanopoulos, Nucl. Phys. B241, 406 (1984);
J. Ellis, C. Kounnas, and D. V. Nanopoulos, Nucl. Phys. B247, 373 (1984).
284. E. Cremmer, S. Ferrara, C. Kounnas, and D. V. Nanopoulos, Phys. Lett. 133B, 61
(1983).
285. G. B. Gelmini, C. Kounnas, and D. V. Nanopoulos, Nucl. Phys. B250, 177 (1985).
286. D. V. Nanopoulos, K. A. Olive, and M. Srednicki, Phys. Lett. 127B, 30 (1983).
287. Ya. B. Zeldovich, Sov. Phys. Usp. 133, 479 (1981).
288. V. Ts. Gurovich and A. A. Starobinsky, Sov. Phys. JETP 50, 844 (1979).
289. Ya. B. Zeldovich, Sov. Astron. Lett. 95, 209 (1981).
290. L. P. Grishchuk and Ya. B. Zeldovich, in: Quantum Structure of Space-Time, edited
by M. Duff and C. J. Isham, Cambridge University Press, Cambridge (1983), p. 353.
291. K. Stelle, Phys. Rev. D16, 953 (1977).
292. A. D. Sakharov, Sov. Phys. JETP 60, 214 (1984).
293. I. Ya. Aref’eva and I. V. Volovich, Phys. Lett. 164B, 287 (1985);
I. Ya. Aref’eva and I. V. Volovich, Theor. Mat. Phys. 64, 866 (1986).
294. M. Reuter and C. Wetterich, Nucl. Phys. B289, 757 (1987).
295. M. D. Pollock, Nucl. Phys. B309, 513 (1988).
296. M. D. Pollock, Phys. Lett. 215B, 635 (1988).
297. J. Ellis, K. Enquist, D. V. Nanopoulos, and M. Quiros, Nucl. Phys. B277, 233
(1986);
P. Binetruy and M. K. Gaillard, Phys. Rev. D34, 3069 (1986);
P. Oh, Phys. Lett. 166B, 292 (1986);
K. Maeda, M. D. Pollock, and C. E. Vayonakis, Class. Quant. Grav. 3, L89 (1986);
S. R. Lonsdale and I. G. Moss, Phys. Lett. 189B, 12 (1987);
M. D. Pollock, Phys. Lett. 199B, 509 (1987);
J. A. Casas and C. Muñoz, Phys. Lett. 216B, 37 (1989);
S. Kalara and K. Olive, Phys. Lett. 218B, 148 (1989).
298. J. A. Wheeler, in: Relativity, Groups, and Topology, edited by C. M. DeWitt and
J. A. Wheeler, Benjamin, New York (1968).
299. B. S. DeWitt, Phys. Rev. 160, 1113 (1967).
300. Quantum Cosmology, edited by L. Z. Fang and R. Ruffini, World Scientific, Singa-
pore (1987).
301. V. N. Ponomarev, A. O. Barvinsky, and Yu. N. Obukhov, Geometrodynamical Meth-
ods and the Gauge Approach to the Theory of Gravitational Interactions [in Russian],
Energoatomizdat, Moscow (1985).
REFERENCES 257

302. J. A. Wheeler, in: Foundational Problems in the Special Sciences, edited by

R. E. Butts and J. Hintikka, D. Reidel, Dordrecht (1977);
also in: Quantum Mechanics, A Half Century Later, edited by J. L. Lopes and
M. Paty, D. Reidel, Dordrecht (1977).
303. H. Everett, Rev. Mod. Phys. 29, 454 (1957).
304. B. S. DeWitt and N. Graham, The Many-Worlds Interpretation of Quantun Me-
chanics, Princeton University Press, Princeton (1973).
305. B. S. DeWitt, Phys. Today 23, 30 (1970);
B. S. DeWitt, Phys. Today 24, 36 (1971).
306. L. Smolin, in: Quantum Theory of Gravity, edited by S. M. Christensen, Adam
Hilger, Bristol (1984).
307. D. Deutsch, Int. J. Theor. Phys. 24, 1 (1985).
308. V. F. Mukhanov, in: Proc. Third Seminar on Quantum Gravity, edited by
M. A. Markov, V. A. Berezin, and V. P. Frolov, World Scientific, Singapore (1984).
309. M. A. Markov and V. F. Mukhanov, Phys. Lett. 127A, 251 (1988).
310. S. W. Hawking, Phys. Rev. D32, 2489 (1985).
311. D. N. Page, Phys. Rev. D32, 2496 (1985).
312. A. D. Sakharov, Sov. Phys. JETP 49, 594 (1979);
A. D. Sakharov, Sov. Phys. JETP 52, 349 (1980).
313. M. A. Markov, Ann. Phys. 155, 333 (1984).
314. J. B. Hartle and S. W. Hawking, Phys. Rev. D28, 2960 (1983).
315. E. P. Tryon, Nature 246, 396 (1973);
P. I. Fomin, Inst. Teor. Fiz. Preprint No. ITF-73-1379;
P. I. Fomin, Dokl. Akad. Nauk SSSR 9, 831 (1975).
316. R. Brout, F. Englert, and E. Gunzig, Ann. Phys. 115, 78 (1978).
317. D. Atkatz and H. Pagels, Phys. Rev. D25, 2065 (1982).
318. A. Vilenkin, Phys. Lett. 117B, 25 (1982).
319. A. D. Linde, Sov. Phys. JETP 60, 211 (1984);
A. D. Linde, Lett. Nuovo Cimento 39, 401 (1984).
320. Ya. B. Zeldovich and A. A. Starobinsky, Sov. Astron. Lett. 10, 135 (1984).
321. V. A. Rubakov, JETP Lett. 39, 107 (1984);
V. A. Rubakov, Phys. Lett. 148B, 280 (1984).
322. A. Vilenkin, Phys. Rev. D30, 509 (1984);
A. Vilenkin, Phys. Rev. D33, 3560 (1986).
323. C. W. Misner, K. S. Thorne, and J. A. Wheeler, Gravitation, W. H. Freeman, San
Francisco (1973).
324. S. W. Hawking and D. N. Page, Nucl. Phys. B264, 185 (1986);
J. J. Halliwell and S. W. Hawking, Phys. Rev. D31, 1777 (1985);
J. J. Halliwell, Phys. Rev. D38, 2468 (1988);
A. Vilenkin, Phys. Rev. D37, 888 (1988);
A. Vilenkin and T. Vachaspati, Phys. Rev. D37, 904 (1988);
L. P. Grishchuk and L. Rozhansky, Phys. Lett. 208B, 369 (1988);
G. W. Gibbons and L. P. Grishchuk, Nucl. Phys. B313, 736 (1989).
REFERENCES 258

325. M. Aryal and A. Vilenkin, Phys. Lett. 199B, 351 (1987).

326. E. Farhi and A. H. Guth, Phys. Lett. 183B, 149 (1987).
327. K. Maeda, K. Sato, M. Sasaki, and H. Kodama, Phys. Lett. 108B, 98 (1982).
328. J. R. Gott, Nature 295, 304 (1982).
329. S. Weinberg, Phys. Rev. Lett. 48, 1776 (1982).
330. P. Ehrenfest, Proc. Amsterdam Acad. 20, 200 (1917).
331. J. D. Barrow and F. J. Tipler, The Anthropic Cosmological Principle, Oxford Uni-
versity Press, Oxford (1986).
332. A. Einstein, B. Podolsky, and N. Rosen, Phys. Rev. 47, 777 (1935).
333. A. D. Linde and M. I. Zelnikov, Phys. Lett. 215B, 59 (1988).
334. A. D. Linde, Phys. Lett. 201B, 437 (1988).
335. R. D. Peccei, J. Sola, and C. Wetterich, Phys. Lett. 195B, 183 (1987).
336. A. D. Linde, Phys. Lett. 211B, 29 (1988).
337. S. Coleman and J. Mandula, Phys. Rev. 159, 1251 (1967).
338. J. Scherk, in: Recent Developments in Gravitation, edited by M. Levy and S. Deser,
Plenum, New York (1979), p. 479;
J. Scherek, in: Geometrical Ideas in Physics [Russian translation], Mir, Moscow
(1983), p. 201;
M. Gell-Mann, Physica Scripta T15, 202 (1987).
339. A. D. Dolgov, in: The Very Early Universe, edited by G. W. Gibbons, S. W. Hawking,
and S. Siklos, Cambridge University Press, Cambridge (1983), p. 449.
340. S. W. Hawking, Phys. Lett. 134B, 403 (1984).
341. T. Banks, Nucl. Phys. B249, 332 (1985).
342. L. Abbott, Phys. Lett. 150B, 427 (1985).
343. S. Barr, Phys. Rev. D36, 1691 (1987).
344. A. D. Linde, Phys. Lett. 200B, 272 (1988).
345. S. Coleman, Nucl. Phys. B307, 867 (1988).
346. S. Coleman, Nucl. Phys. B310, 643 (1988).
347. T. Banks, Nucl. Phys. B309, 493 (1988).
348. S. Weinberg, Phys. Rev. Lett. 59, 2607 (1987);
Rev. Mod. Phys. 61, 1 (1989).
349. S. Giddings and A. Strominger, Nucl. Phys. B307, 854 (1988).
350. S. W. Hawking, Phys. Lett. 195B, 337 (1987);
S. W. Hawking, Phys. Rev. D37, 904 (1988).
351. G. V. Lavrelashvili, V. A. Rubakov, and P. G. Tinyakov, JETP Lett. 46, 167 (1987);
G. V. Lavrelashvili, V. A. Rubakov, and P. G. Tinyakov, Nucl. Phys. B299, 757
(1988).
352. S. Giddings and A. Strominger, Nucl. Phys. B306, 890 (1988).
353. I. Antoniadis, C. Bachas, J. Ellis, and D. V. Nanopoulos, Phys. Lett. 211B, 393
(1988).
354. E. Tomboulis, preprint UCLA/89/TER/8 (1989).
355. A. D. Linde, in: Proceedings of XXIV International Conference on High Energy
Physics, Munich 1988 (Springer-Verlag, Berlin 1989), p. 357.
REFERENCES 259

356. J. Polchinski, Phys. Lett. 219B, 251 (1989).

357. W. Fischer, I. Klebanov, J. Polchinski, and L. Susskind, Nucl. Phys. B327, 157
(1989).
358. V. A. Rubakov and M. E. Shaposhnikov, Mod. Phys. Lett. 4A, 107 (1989).
359. A. D. Linde, Phys. Lett. 227B, 352 (1989).
360. P. B. Arnold, Phys. Rev. D40, 613 (1989);
M. Sher, Phys. Rep. C179, 274 (1989).
361. A. D. Linde, D. H. Lyth, Phys. Lett. B246, 353 (1990).
362. A. D. Linde, Stanford University preprint SU-ITP-883 (1991) to be published in
Phys. Lett.